CN113065573A - User classification method, user classification device and electronic equipment - Google Patents
User classification method, user classification device and electronic equipment Download PDFInfo
- Publication number
- CN113065573A CN113065573A CN202010001309.8A CN202010001309A CN113065573A CN 113065573 A CN113065573 A CN 113065573A CN 202010001309 A CN202010001309 A CN 202010001309A CN 113065573 A CN113065573 A CN 113065573A
- Authority
- CN
- China
- Prior art keywords
- user
- relation
- feature
- individual
- attention
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 210
- 239000013598 vector Substances 0.000 claims abstract description 404
- 230000007246 mechanism Effects 0.000 claims abstract description 166
- 239000011159 matrix material Substances 0.000 claims description 12
- 238000004590 computer program Methods 0.000 claims description 11
- 230000009466 transformation Effects 0.000 claims description 11
- 238000013507 mapping Methods 0.000 claims description 7
- 238000010586 diagram Methods 0.000 description 18
- 230000006870 function Effects 0.000 description 7
- 230000003252 repetitive effect Effects 0.000 description 4
- 238000012545 processing Methods 0.000 description 3
- 238000013473 artificial intelligence Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000004927 fusion Effects 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 238000007792 addition Methods 0.000 description 1
- 230000004075 alteration Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000002452 interceptive effect Effects 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 238000003058 natural language processing Methods 0.000 description 1
- 239000013307 optical fiber Substances 0.000 description 1
- 230000001737 promoting effect Effects 0.000 description 1
- 238000005295 random walk Methods 0.000 description 1
- 238000005215 recombination Methods 0.000 description 1
- 230000006798 recombination Effects 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 230000001131 transforming effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/24317—Piecewise classification, i.e. whereby each classification requires several discriminant rules
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
A user classification method, a user classification device and an electronic device are disclosed. The user classification method comprises the following steps: acquiring a first individual characteristic of a target user; traversing by using a meta-path method aiming at different relation types of the relation users of the target user to obtain a plurality of meta-paths each containing a plurality of relation users; acquiring a second characteristic and a relation characteristic of the relation user in each meta path; using at least one attention mechanism based on the first individual feature, the second individual feature and the relationship feature of the stitching to obtain an attention feature vector; and obtaining a classification result of the target user based on the attention feature vector using a multi-layer perceptron network. Thus, the accuracy, recall rate and robustness of user classification are improved.
Description
Technical Field
The present application relates to the field of artificial intelligence technology, and more particularly, to a user classification method, a user classification apparatus, and an electronic device.
Background
User classification refers to classifying users into different user groups, for example, users may be classified according to their attributes, such as age, gender, income, education, occupation, family, regional factors, and the like, of their personal attribute characteristics.
With the increase of user data and the development of machine learning technology, user classification presents the following several trends. First, there are increasing dimensions for classifying users, i.e., there are more and more scenes involved in user classification, such as overdue identification of financial products by users. Secondly, the user classification becomes less and less time-consuming, which also makes the requirements for the user data used for classification higher and requires improved user classification techniques. Third, user classification has evolved from artificial classification to artificial intelligence classification based on user big data, i.e., user classification by extracting features from the user's data using a machine learning algorithm.
Accordingly, it is desirable to provide an improved user classification scheme.
Disclosure of Invention
The present application is proposed to solve the above-mentioned technical problems. The embodiment of the application provides a user classification method, a user classification device and electronic equipment, which combine individual features of a target user and individual features and relationship features of a relationship user of the target user and use an attention mechanism to fuse the individual features and the relationship features, so that the accuracy, recall rate and robustness of user classification are improved.
According to an aspect of the present application, there is provided a user classification method, including: acquiring a first individual feature of a target user, wherein the first individual feature is used for representing an individual attribute of the target user; traversing by using a meta-path method aiming at different relation types of the relation users of the target user to obtain a plurality of meta-paths each containing a plurality of relation users, wherein the meta-path method is used for obtaining the plurality of relation users based on similarity of different relation types in the relation users through similarity search; acquiring a second individual characteristic and a relation characteristic of a relation user in each meta path, wherein the second individual characteristic is used for representing individual attributes of the relation user, and the relation characteristic is used for representing the relation between the relation user and the target user; using at least one attention mechanism for calculating degrees of importance of different features to the target user based on the stitched first individual feature, the second individual feature and the relationship feature to obtain an attention feature vector, and the attention feature vector is a feature vector in which different features have different weights based on the degrees of importance; and obtaining a classification result of the target user based on the attention feature vector by using a multi-layer perceptron network, wherein the multi-layer perceptron network is used for classifying based on the feature vector.
In the user classification method, the meta path has a corresponding associated relationship type.
In the above user classification method, obtaining an attention feature vector using at least one attention mechanism based on the first individual feature, the second individual feature and the relationship feature of the concatenation comprises: for each meta path, splicing the second characteristic and the relation characteristic of each relation user in the plurality of relation users to obtain a plurality of relation user characteristics, wherein the relation user characteristics are the spliced second characteristic and the relation characteristic; splicing the plurality of relational user features with the first individual feature to obtain a plurality of first splicing features, wherein the first splicing features are the spliced first individual feature, second individual feature and splicing feature; using at least one attention mechanism based on the first splicing features to obtain a plurality of sub-attention vectors corresponding to each meta-path, wherein the sub-attention vectors are feature vectors used for representing second feature of the corresponding relation user and importance degree of the relation feature relative to the target user corresponding to each meta-path; and concatenating the plurality of sub-attention vectors to obtain the attention feature vector.
In the user classification method, for each meta path, the obtaining a plurality of relational user features by concatenating the second feature and the relational feature of each user of the plurality of users includes: for each meta path, mapping the respective second individual characteristic and relationship characteristic of each user in the plurality of users into a second individual characteristic vector and relationship characteristic vector with the same length through a transformation matrix, wherein the transformation matrix is used for converting different vectors into the same length; and splicing the second individual feature vector and the relation feature vector with the same length into the plurality of relation user features.
In the above user classification method, using at least one attention mechanism based on the plurality of stitched features to obtain a plurality of sub-attention vectors corresponding to each meta-path comprises: using a node attention mechanism based on the plurality of first stitching features to obtain a node attention vector, the node attention mechanism being used to calculate degrees of importance of different users as nodes to the target user, and the node attention vector being a feature vector in which features of different users have different weights based on the degrees of importance; obtaining a plurality of node features from the node attention vector and the each relational user feature; and obtaining the sub-attention vector based on the sum of the plurality of node features.
In the above user classification method, obtaining the sub-attention vector based on the sum of the plurality of node features comprises: for each meta-path, obtaining a sum feature vector from the plurality of node features; stitching the sum feature vector with the first individual feature to obtain a second stitched vector; using a feature attention mechanism based on the second stitching vector to obtain a feature attention vector, the feature attention mechanism for calculating degrees of importance of different types of features to the target user, and the feature attention vector being a feature vector in which different types of features have different weights based on the degrees of importance; obtaining a user set feature vector from the feature attention vector and the second stitching vector; and obtaining the sub-attention vector based on a plurality of user set feature vectors corresponding to a plurality of meta-paths.
In the above user classification method, obtaining the sub-attention vector based on a plurality of user set feature vectors corresponding to a plurality of meta-paths includes: stitching the plurality of user set feature vectors with the first individual feature respectively to obtain a plurality of third stitching vectors; using a semantic attention mechanism based on the third stitching vector to obtain a plurality of semantic attention vectors, the semantic attention mechanism for calculating degrees of importance of different relationship types to the target user, and the feature attention vectors being feature vectors in which features of different relationship types have different weights based on the degrees of importance; and obtaining the sub-attention vector from each corresponding user set feature vector and semantic attention vector.
In the user classification method, the target user includes one of a user account, user equipment, and a user item.
In the user classification method, the different relationship types include one of a social relationship, a fund relationship, and a device login relationship.
In the user classification method, the classification result includes one of a financial overdue result, a search recommendation result, and a false transaction result.
According to another aspect of the present application, there is provided a financial user overdue identification method, including: acquiring a first individual feature of a financial user, wherein the first individual feature is used for representing an individual attribute of the financial user; traversing by using a meta-path method aiming at different relation types of relation users of the financial users to acquire a plurality of meta-paths each including a plurality of relation users, wherein the meta-path method is used for acquiring the plurality of relation users similar based on different relation types in the relation users through similarity search; acquiring a second physical characteristic and a relation characteristic of the relation user in each meta-path, wherein the second physical characteristic is used for representing individual attributes of the relation user, and the relation characteristic is used for representing the relation between the relation user and the financial user; using at least one attention mechanism for calculating degrees of importance of different features to the target user based on the stitched first individual feature, the second individual feature and the relationship feature to obtain an attention feature vector, and the attention feature vector is a feature vector in which different features have different weights based on the degrees of importance; and identifying whether the financial user is overdue based on the attention feature vector by using a multi-layer sensor network, wherein the multi-layer sensor network is used for classifying based on the feature vector and is provided with a label used for representing the overdue of the financial user.
According to still another aspect of the present application, there is provided a search recommendation method for a target user, including: acquiring a first individual feature of a target user, wherein the first individual feature is used for representing an individual attribute of the target user; traversing by using a meta-path method aiming at different relation types of the relation users of the target user to obtain a plurality of meta-paths each containing a plurality of relation users, wherein the meta-path method is used for obtaining the plurality of relation users based on similarity of different relation types in the relation users through similarity search; acquiring a second individual characteristic and a relation characteristic of a relation user in each meta path, wherein the second individual characteristic is used for representing individual attributes of the relation user, and the relation characteristic is used for representing the relation between the relation user and the target user; using at least one attention mechanism for calculating degrees of importance of different features to the target user based on the stitched first individual feature, the second individual feature and the relationship feature to obtain an attention feature vector, and the attention feature vector is a feature vector in which different features have different weights based on the degrees of importance; classifying the target user based on the attention feature vector by using a multi-layer perceptron network, wherein the multi-layer perceptron network is used for classifying based on the feature vector and is provided with a label for representing a category recommended to the target user for searching; and carrying out search recommendation to the target user according to the category.
According to yet another aspect of the present application, there is provided a false transaction identification method, including: acquiring a first individual characteristic of a transaction user, wherein the first individual characteristic is used for representing an individual attribute of the financial user; traversing by using a meta-path method aiming at different relationship types of relationship users of the transaction user to obtain a plurality of meta-paths each containing a plurality of relationship users, wherein the meta-path method is used for obtaining the plurality of relationship users based on similarity of different relationship types in the relationship users through similarity search; acquiring a second physical characteristic and a relation characteristic of a relation user in each meta-path, wherein the second physical characteristic is used for representing individual attributes of the relation user, and the relation characteristic is used for representing the relation between the relation user and the transaction user; using at least one attention mechanism for calculating degrees of importance of different features to the target user based on the stitched first individual feature, the second individual feature and the relationship feature to obtain an attention feature vector, and the attention feature vector is a feature vector in which different features have different weights based on the degrees of importance; and identifying whether the transaction user carries out false transaction based on the attention feature vector by using a multi-layer perceptron network, wherein the multi-layer perceptron network is used for carrying out classification based on the feature vector and is provided with a label used for representing false transaction.
According to still another aspect of the present application, there is provided a method of identifying a user's intention to receive a coupon, including: acquiring a first individual characteristic of a user receiving a coupon, wherein the first individual characteristic is used for representing an individual attribute of the user; traversing by using a meta-path method aiming at different relation types of relation users of the users to acquire a plurality of meta-paths each including a plurality of relation users, wherein the meta-path method is used for acquiring the plurality of relation users similar based on different relation types in the relation users through similarity search; acquiring a second individual characteristic and a relation characteristic of a relation user in each meta path, wherein the second individual characteristic is used for representing individual attributes of the relation user, and the relation characteristic is used for representing the relation between the relation user and a user receiving the coupons; using at least one attention mechanism for calculating degrees of importance of different features to the target user based on the stitched first individual feature, the second individual feature and the relationship feature to obtain an attention feature vector, and the attention feature vector is a feature vector in which different features have different weights based on the degrees of importance; and identifying whether the coupon picked up by the user is malicious intention or not based on the attention feature vector by using a multi-layer perceptron network, wherein the multi-layer perceptron network is used for classifying based on the feature vector and is provided with a label used for representing the malicious intention of the user.
According to another aspect of the present application, there is provided a user classification apparatus including: the target feature acquiring unit is used for acquiring a first individual feature of a target user, wherein the first individual feature is used for representing an individual attribute of the target user; the meta-path traversing unit is used for traversing by using a meta-path method aiming at different relation types of the relation users of the target user to acquire a plurality of meta-paths each containing a plurality of relation users, and the meta-path method is used for acquiring the plurality of relation users similar based on different relation types in the relation users through similarity search; a relation feature obtaining unit, configured to obtain a second feature and a relation feature of the relation user in each meta path, where the second feature is used to represent an individual attribute of the relation user, and the relation feature is used to represent a relation between the relation user and the target user; an attention mechanism unit for using at least one attention mechanism based on the first individual feature, the second individual feature and the relationship feature of the stitching to obtain an attention feature vector, the at least one attention mechanism being used for calculating degrees of importance of different features to the target user, and the attention feature vector being a feature vector in which different features have different weights based on the degrees of importance; and a classification result obtaining unit configured to obtain a classification result of the target user based on the attention feature vector using a multi-layered perceptron network, the multi-layered perceptron network configured to perform classification based on the feature vector.
In the above user classification apparatus, the meta path has a corresponding associated relationship type.
In the above user classification apparatus, the attention mechanism unit is configured to: for each meta path, splicing the second characteristic and the relation characteristic of each relation user in the plurality of relation users to obtain a plurality of relation user characteristics, wherein the relation user characteristics are the spliced second characteristic and the relation characteristic; splicing the plurality of relational user features with the first individual feature to obtain a plurality of first splicing features, wherein the first splicing features are the spliced first individual feature, second individual feature and splicing feature; using at least one attention mechanism based on the first splicing features to obtain a plurality of sub-attention vectors corresponding to each meta-path, wherein the sub-attention vectors are feature vectors used for representing second feature of the corresponding relation user and importance degree of the relation feature relative to the target user corresponding to each meta-path; and concatenating the plurality of sub-attention vectors to obtain the attention feature vector.
In the foregoing user classification device, the attention mechanism unit, for each meta path, concatenating the second characteristic feature and the relationship feature of each user of the multiple users to obtain multiple relationship user characteristics includes: for each meta path, mapping the respective second individual characteristic and relationship characteristic of each user in the plurality of users into a second individual characteristic vector and relationship characteristic vector with the same length through a transformation matrix, wherein the transformation matrix is used for converting different vectors into the same length; and splicing the second individual feature vector and the relation feature vector with the same length into the plurality of relation user features.
In the above user classification apparatus, the attention mechanism unit using at least one attention mechanism based on the plurality of stitched features to obtain a plurality of sub-attention vectors corresponding to each meta-path includes: using a node attention mechanism based on the plurality of first stitching features to obtain a node attention vector, the node attention mechanism being used to calculate degrees of importance of different users as nodes to the target user, and the node attention vector being a feature vector in which features of different users have different weights based on the degrees of importance; obtaining a plurality of node features from the node attention vector and the each relational user feature; and obtaining the sub-attention vector based on the sum of the plurality of node features.
In the above-described user classification device, the obtaining, by the attention mechanism unit, the sub-attention vector based on the sum of the plurality of node features may include: for each meta-path, obtaining a sum feature vector from the plurality of node features; stitching the sum feature vector with the first individual feature to obtain a second stitched vector; using a feature attention mechanism based on the second stitching vector to obtain a feature attention vector, the feature attention mechanism for calculating degrees of importance of different types of features to the target user, and the feature attention vector being a feature vector in which different types of features have different weights based on the degrees of importance; obtaining a user set feature vector from the feature attention vector and the second stitching vector; and obtaining the sub-attention vector based on a plurality of user set feature vectors corresponding to a plurality of meta-paths.
In the above-described user classification device, the obtaining, by the attention mechanism unit, the sub-attention vector based on a plurality of user set feature vectors corresponding to a plurality of meta paths includes: stitching the plurality of user set feature vectors with the first individual feature respectively to obtain a plurality of third stitching vectors; using a semantic attention mechanism based on the third stitching vector to obtain a plurality of semantic attention vectors, the semantic attention mechanism for calculating degrees of importance of different relationship types to the target user, and the feature attention vectors being feature vectors in which features of different relationship types have different weights based on the degrees of importance; and obtaining the sub-attention vector from each corresponding user set feature vector and semantic attention vector.
In the user classification apparatus, the target user includes one of a user account, user equipment, and a user item.
In the user classification apparatus, the different relationship types include one of a social relationship, a fund relationship, and a device login relationship.
In the user classification device, the classification result includes one of a financial overdue result, a search recommendation result, and a false transaction result.
According to still another aspect of the present application, there is provided a financial user overdue identification apparatus including: the financial user characteristic acquisition unit is used for acquiring a first individual characteristic of a financial user, and the first individual characteristic is used for representing an individual attribute of the financial user; the financial meta-path traversing unit is used for traversing by using a meta-path method aiming at different relationship types of the relationship users of the financial users to acquire a plurality of meta-paths each containing a plurality of relationship users, and the meta-path method is used for acquiring the plurality of relationship users similar based on different relationship types in the relationship users through similarity search; the financial relation characteristic acquisition unit is used for acquiring a second individual characteristic and a relation characteristic of the relation user in each meta path, wherein the second individual characteristic is used for representing individual attributes of the relation user, and the relation characteristic is used for representing the relation between the relation user and the financial user; a financial attention mechanism unit for using at least one attention mechanism based on the first individual feature, the second individual feature and the relationship feature of the concatenation to obtain an attention feature vector, the at least one attention mechanism being used to calculate degrees of importance of different features to the target user, and the attention feature vector being a feature vector in which different features have different weights based on the degrees of importance; and the financial user overdue identification unit is used for identifying whether the financial user is overdue or not based on the attention characteristic vector by using a multi-layer sensor network, the multi-layer sensor network is used for classifying based on the characteristic vector and is provided with a label used for representing the overdue of the financial user.
According to still another aspect of the present application, there is provided a search recommendation apparatus for a target user, including: the search user characteristic acquisition unit is used for acquiring a first individual characteristic of a target user, and the first individual characteristic is used for representing an individual attribute of the target user; a search meta-path traversal unit, configured to traverse, using a meta-path method, for different relationship types of a relationship user of the target user, to obtain multiple meta-paths each including multiple relationship users, where the meta-path method is used to obtain, through similarity search, the multiple relationship users that are similar based on different relationship types among the relationship users; a search relationship feature obtaining unit, configured to obtain a second feature and a relationship feature of the relationship user in each meta path, where the second feature is used to represent an individual attribute of the relationship user, and the relationship feature is used to represent a relationship between the relationship user and the target user; a search attention mechanism unit for using at least one attention mechanism based on the first individual feature, the second individual feature and the relationship feature of the concatenation to obtain an attention feature vector, the at least one attention mechanism being used for calculating degrees of importance of different features to the target user, and the attention feature vector being a feature vector in which different features have different weights based on the degrees of importance; the user classification unit is used for classifying the target user based on the attention feature vector by using a multilayer perceptron network, the multilayer perceptron network is used for classifying based on the feature vector and is provided with a label for expressing a category recommended to the target user for searching; and the search recommending unit is used for carrying out search recommendation on the target user according to the category.
According to yet another aspect of the present application, there is provided a false transaction identifying device comprising: the transaction user characteristic acquisition unit is used for acquiring a first individual characteristic of a transaction user, and the first individual characteristic is used for representing an individual attribute of the financial user; the transaction meta-path traversing unit is used for traversing by using a meta-path method aiming at different relationship types of relationship users of the transaction users to acquire a plurality of meta-paths each including a plurality of relationship users, and the meta-path method is used for acquiring the plurality of relationship users similar based on different relationship types in the relationship users through similarity search; the transaction relationship characteristic acquisition unit is used for acquiring a second characteristic and a relationship characteristic of the relationship user in each meta path, wherein the second characteristic is used for representing the individual attribute of the relationship user, and the relationship characteristic is used for representing the relationship between the relationship user and the transaction user; a transaction attention mechanism unit for using at least one attention mechanism based on the first individual feature, the second individual feature and the relationship feature of the concatenation to obtain an attention feature vector, the at least one attention mechanism being used to calculate degrees of importance of different features to the target user, and the attention feature vector being a feature vector in which different features have different weights based on the degrees of importance; and the false transaction identification unit is used for identifying whether the transaction user carries out false transaction or not based on the attention characteristic vector by using a multi-layer sensor network, and the multi-layer sensor network is used for carrying out classification based on the characteristic vector and is provided with a label for representing false transaction.
According to still another aspect of the present application, there is provided an apparatus for recognizing a user's intention to take a coupon, including: an intention user characteristic obtaining unit, configured to obtain a first individual characteristic of a user who picks up a coupon, the first individual characteristic being used for representing an individual attribute of the user; the intention meta path traversing unit is used for traversing by using a meta path method aiming at different relation types of relation users of the users to acquire a plurality of meta paths each containing a plurality of relation users, and the meta path method is used for acquiring the plurality of relation users similar based on different relation types in the relation users through similarity search; an intention relation characteristic obtaining unit, configured to obtain a second individual characteristic and a relation characteristic of the relation user in each meta path, where the second individual characteristic is used to represent an individual attribute of the relation user, and the relation characteristic is used to represent a relation between the relation user and the user receiving the coupon; an intent attention mechanism unit for using at least one attention mechanism based on the first individual feature, the second individual feature and the relationship feature of the stitching to obtain an attention feature vector, the at least one attention mechanism being used to calculate degrees of importance of different features to the target user, and the attention feature vector being a feature vector in which different features have different weights based on the degrees of importance; and a malicious intention identification unit, which is used for identifying whether the coupon picked up by the user is malicious intention or not based on the attention feature vector by using a multi-layer perceptron network, wherein the multi-layer perceptron network is used for classifying based on the feature vector and is provided with a label used for representing the malicious intention of the user.
According to still another aspect of the present application, there is provided an electronic apparatus including: a processor; and a memory in which are stored computer program instructions that, when executed by the processor, cause the processor to perform the user classification method, the financial user overdue identification method, the search recommendation method of the target user, the false transaction identification method, and the identification method of the user's intention to receive the coupon as described above.
According to yet another aspect of the present application, there is provided a computer readable medium having stored thereon computer program instructions which, when executed by a processor, cause the processor to perform the user classification method, the financial user overdue identification method, the search recommendation method of the target user, the false transaction identification method, and the identification method of the user's intention to receive a coupon as described above.
The user classification method, the user classification device and the electronic equipment can combine the individual features of the target user and the individual features and the relation features of the relation user of the target user, and use the attention mechanism to fuse the individual features and the relation features, so that the relation features of the relation user of the target user are combined on the basis of the individual features of the target user, the users can be classified based on the combination of the individual features and the relation features, and the accuracy of user classification is improved.
In addition, by using the meta-path method, the unknown topological relation structure of the target user and the relation user can be adaptively mined on the basis of the relation characteristics, and the individual characteristics of the target user and the relation user are fused, so that the recall rate of user classification is improved.
In addition, the attention mechanism can realize the capture of the importance degree of the individuals and the relations while fusing the individual characteristics and the relation characteristics, so that the accuracy of user classification can be improved, and the robustness of user classification can be improved as the importance degree of the characteristics can be distinguished during classification to emphasize important characteristics more and ignore non-important characteristics.
Drawings
The above and other objects, features and advantages of the present application will become more apparent by describing in more detail embodiments of the present application with reference to the attached drawings. The accompanying drawings are included to provide a further understanding of the embodiments of the application and are incorporated in and constitute a part of this specification, illustrate embodiments of the application and together with the description serve to explain the principles of the application. In the drawings, like reference numbers generally represent like parts or steps.
Fig. 1 is a schematic diagram illustrating an application scenario of a user classification method according to an embodiment of the present application.
Fig. 2 illustrates a flow chart of a user classification method according to an embodiment of the application.
Fig. 3 illustrates a schematic diagram of a network architecture to which the user classification method according to the embodiment of the present application is applied.
FIG. 4 illustrates a flow chart of a financial user overdue identification method according to an embodiment of the application.
Fig. 5 illustrates a flowchart of a search recommendation method of a target user according to an embodiment of the present application.
Fig. 6 illustrates a flow chart of a false transaction identification method according to an embodiment of the application.
FIG. 7 illustrates a flow chart of a method of identifying a user's intent to pick up a coupon according to an embodiment of the application.
Fig. 8 illustrates a block diagram of a user classification device according to an embodiment of the application.
Fig. 9 illustrates a block diagram of a financial user overdue identification apparatus according to an embodiment of the present application.
Fig. 10 illustrates a block diagram of a search recommendation apparatus of a target user according to an embodiment of the present application.
FIG. 11 illustrates a block diagram of a false transaction identification device according to an embodiment of the present application.
Fig. 12 illustrates a block diagram of an apparatus for recognizing a user's intention to take a coupon according to an embodiment of the present application.
FIG. 13 illustrates a block diagram of an electronic device in accordance with an embodiment of the present application.
Detailed Description
Hereinafter, example embodiments according to the present application will be described in detail with reference to the accompanying drawings. It should be understood that the described embodiments are only some embodiments of the present application and not all embodiments of the present application, and that the present application is not limited by the example embodiments described herein.
Summary of the application
As described above, as the number of user classification application scenarios increases, more and more user data are used to implement the user classification task. For example, for overdue identification of a user's financial product, whether the user is overdue is classified by using individual characteristics (such as purchase amount, age, etc.) and relationship characteristics (such as login to a device, social relationship, etc.) of the user.
The current common methods for financial overdue classification are roughly divided into two directions: the method is based on individual features, such as GBDT for realizing classification according to the individual features; and a method based on the relation characteristics, such as label propagation, random walk, subgraph mining and the like, is used for realizing classification.
However, the first method only considers individual features and ignores the topology information of the network, and the second method cannot adaptively mine unknown topology and fuse individual features, so that the accuracy and recall rate of the common methods are low.
In view of the above technical problem, the basic idea of the present application is to fuse individual and relational characteristics of users through a heterogeneous network based on at least one attention mechanism.
Specifically, the user classification method, the user classification device, and the electronic device provided by the present application first obtain a first individual feature of a target user, where the first individual feature is used to represent an individual attribute of the target user, then traverse through a meta-path method for different relationship types of a relationship user of the target user to obtain multiple meta-paths each including multiple relationship users, where the meta-path method is used to obtain the multiple relationship users similar based on different relationship types in the relationship users through similarity search, and then obtain a second individual feature and a relationship feature of the relationship user in each meta-path, where the second individual feature is used to represent an individual attribute of the relationship user, and the relationship feature is used to represent a relationship between the relationship user and the target user, and then based on the spliced first individual feature, The second individual feature and the relationship feature use at least one attention mechanism for calculating degrees of importance of different features to the target user to obtain an attention feature vector, and the attention feature vector is a feature vector in which different features have different weights based on the degrees of importance, and finally a classification result of the target user is obtained based on the attention feature vector using a multi-layered perceptron network for classification based on feature vectors.
Therefore, the user classification method, the user classification device and the electronic equipment provided by the application can combine the individual features of the target user and the individual features and the relationship features of the relationship user thereof, and use the attention mechanism to fuse the individual features and the relationship features, so that the relationship features of the relationship user of the target user are combined on the basis of the individual features of the target user, the user can be classified based on the combination of the individual features and the relationship features, and the accuracy of user classification is improved.
In addition, by using the meta-path method, the unknown topological relation structure of the target user and the relation user can be adaptively mined on the basis of the relation characteristics, and the individual characteristics of the target user and the relation user are fused, so that the recall rate of user classification is improved.
In addition, the attention mechanism can realize the capture of the importance degree of the individuals and the relations while fusing the individual characteristics and the relation characteristics, so that the accuracy of user classification can be improved, and the robustness of user classification can be improved as the importance degree of the characteristics can be distinguished during classification to emphasize important characteristics more and ignore non-important characteristics.
Fig. 1 is a schematic diagram illustrating an application scenario of a user classification method according to an embodiment of the present application. As shown in fig. 1, individual and relational characteristics of a target user T and its plurality of relational users A, B, C, … are obtained by a cloud server C. Then, the identification result TR of the target user T is acquired by the server apparatus F for user identification. Here, the server apparatus F includes a heterogeneous network based on at least one attention mechanism, which further includes a module applying a meta path method, a module applying an attention mechanism, and a module of a multi-layered perceptron network. Finally, the recognition result TR may be transmitted from the server device F to another terminal, such as the mobile terminal M, which needs the information of the target user T, and presented to the other user U, who needs the information of the target user T, through the mobile terminal M.
Having described the general principles of the present application, various non-limiting embodiments of the present application will now be described with reference to the accompanying drawings.
Exemplary method one
Fig. 2 illustrates a flow chart of a user classification method according to an embodiment of the application.
As shown in fig. 2, the user classification method according to the embodiment of the present application includes the following steps.
Step S110, obtaining a first individual feature of a target user, wherein the first individual feature is used for representing an individual attribute of the target user. Here, the target user is a user who needs to be classified, for example, to recognize whether the user has overdue of a financial product. The first individual characteristic of the target user can be age, occupation, income and other various data reflecting individual attributes of the user. Also, as will be appreciated by those skilled in the art, the first individual feature of the target user may correspond to different personal data based on the specific classification scenario of the target user.
In addition, in the embodiment of the present application, the target user may be multiple types of users, that is, the target user is not limited to a natural person or a legal person, and may also include other types of users such as a user account, user equipment, and a user item. For example, in the embodiment of the present application, the target user may be a mobile terminal, such as a personal payment device used in a store, for example, a POS, and the first feature data thereof will include transaction feature data, fund feature data, and the like.
Step S120, for different relationship types of the relationship user of the target user, using a meta-path method to traverse to obtain multiple meta-paths each including multiple relationship users, where the meta-path method is used to obtain, through similarity search, the multiple relationship users similar based on different relationship types in the relationship users. As described above, in the embodiment of the present application, a relationship user of the target user is also involved in addition to the individual feature of the target user. Here, the relationship user of the target user is, for example, a relationship user having a social relationship with the target user, a relationship user having a fund relationship with the target user, or a relationship user logged in using the same device as the target user.
In addition, as described above, in the embodiment of the present application, the target user may be various types of users, and may also include other identifiers that may represent people, such as a user account, or a device or an article used by a person, in addition to the user itself as a person. Accordingly, the target user's relationship user may also be a different type of user, such as a user account, user equipment, or user item, etc. Moreover, the target user's relationship users are not limited to one type of user, but may also include multiple types of users, for example, several relationship users are users themselves, and another relationship user is a user account, a user device, a user item, and the like.
Since the relationship users in the embodiment of the present application are relationship users of different relationship types of the target user, and may also be users of different types, the user set composed of the relationship users according to the embodiment of the present application is substantially a heterogeneous user set. Thus, in embodiments of the present application, a set of relationship users corresponding to different relationship types is generated by using a meta-path method traversal. Here, the meta-path method is substantially a method for determining a subset of a target set through similarity, and in this embodiment of the application, the meta-path method is used for obtaining, through similarity search, the plurality of relationship users that are similar based on different relationship types among the relationship users.
For example, as shown in fig. 1, the relationship users of a certain target user T include a user a, a user B, a user C, a user D, and a user E, and the relationship types include both social relationships and fund relationships. And traversing by using a meta path method to obtain a meta path corresponding to the social relationship, namely user A, user C and user D, and a meta path corresponding to the fund relationship, namely user B, user E and user D.
Thus, for a set of relational users, a meta-path method may be used to traverse to produce multiple meta-paths corresponding to different relationship types, such as meta-path 1 through meta-path P as shown in FIG. 2. Here, fig. 2 illustrates a schematic diagram of a network architecture to which the user classification method according to the embodiment of the present application is applied. Also, one or more meta paths may be generated for a certain relationship type. For example, in meta-path 1 to meta-path P shown in fig. 2, one or more meta-paths may correspond to a social relationship, or one or more meta-paths may correspond to a fund relationship.
Step S130, obtaining a second individual feature and a relation feature of the relation user in each meta-path, wherein the second individual feature is used for representing an individual attribute of the relation user, and the relation feature is used for representing a relation between the relation user and the target user. That is, for the determined plurality of meta-paths, a second characteristic feature and a relational feature of each of the plurality of users included in each meta-path are determined. For example, for the meta path { user a, user C, user D } as described above, the second individual characteristics of user a, user C, and user D are determined, and then the relationship characteristics of user a, user C, and user D are determined.
As described above, the second individual feature may be data reflecting individual attributes of the user, such as age, occupation, and the like, depending on the classification scenario to be performed, similarly to the first individual feature. The relationship characteristic of the user may be represented by a path including the user in the meta path, for example, a path in which the relationship characteristic of the user C is a → C, a path in which the relationship characteristic of the user D is C → D, or the like. In addition, for the first user in each meta path, for example, the user a as described above, a virtual path from a certain user to the user a may also be constructed, thereby serving as the relationship characteristic of the user a.
In this way, for the generated multiple meta paths, a user set and a relationship set corresponding to the multiple meta paths are actually obtained. For example, the user set is the user set shown in FIG. 2 The relationship set is the relationship set shown in FIG. 2And, the user setAnd the relation setAre corresponding.
That is, in the user classification method according to the embodiment of the present application, the meta path has a corresponding associated relationship type.
Step S140, using at least one attention mechanism based on the first individual feature, the second individual feature and the relationship feature of the concatenation to obtain an attention feature vector, wherein the at least one attention mechanism is used for calculating importance degrees of different features to the target user, and the attention feature vector is a feature vector in which different features have different weights based on the importance degrees.
As described above, by obtaining the first individual feature of the target user, the second individual feature of the relational user, and the relational feature, the feature vector finally used for user classification can be obtained by fusing the above features. In the embodiment of the present application, in order to distinguish the importance degree for the target user, an attention mechanism is adopted to fuse the first individual feature, the second individual feature, and the relationship feature.
Here, Attention mechanism (Attention) utilizes the principle of Attention similar to human, by focusing on a portion that needs a great deal of Attention and then putting more Attention into this portion to acquire more detailed information of a target that needs Attention, while suppressing other useless information. The method can be applied to various deep learning tasks of different types, such as natural language processing, image recognition, voice recognition and the like, so as to select information which is more critical to the current task target from a plurality of information.
In the embodiment of the present application, the second individual characteristic and the relationship characteristic relate to different characteristics of different relationship users having different relationship types, that is, each relationship, each user and each characteristic may have different degrees of importance for the classification of the target user, and therefore, different attention mechanisms may be adopted to calculate the degrees of importance of the second individual characteristic and the relationship characteristic to the target user respectively.
Also, for each relationship, each user, and each feature, one or more attention mechanisms may be used for one or more of them to calculate their importance to the target user, and thus, in the present embodiment, one or more attention mechanisms may be used to capture features that are relatively important to the classification result from one or more aspects.
Thus, by using at least one attention mechanism based on the stitched first individual feature, the second individual feature and the relationship feature, the obtained attention feature vector will give more weight to features having a greater degree of importance to the classification result of the target user, thereby facilitating the accuracy of user classification.
The attention mechanism employed in the embodiments of the present application will be specifically described below.
And S150, obtaining a classification result of the target user based on the attention feature vector by using a multilayer perceptron network, wherein the multilayer perceptron network is used for classifying based on the feature vector. Here, the classification result may be a financial overdue result, a search recommendation result, a false transaction result, or the like depending on a classification scenario, such as a financial transaction scenario, a recommendation transaction scenario, an e-commerce transaction scenario, or the like, to which the user classification method according to the embodiment of the present application is applied, which will be described in further detail below.
As described above, since the attention feature vector gives different weights to different features, different degrees of importance of the respective features to the user classification are distinguished. Therefore, more accurate user classification results can be obtained through the multilayer perceptron network.
In addition, as described above, according to the user classification method of the embodiment of the present application, the topology information of the relational user network of the target user is effectively utilized through the meta-path method, so that the unknown topology structure and the fusion individual features can be adaptively mined, and the recall rate of user classification is improved.
In addition, due to the effective fusion of the individual characteristics and the relationship characteristics, the robustness of user classification is also improved.
Attention mechanism
In the following, the attention mechanism will be explained in connection with a schematic network architecture of a user classification method according to an embodiment of the present application as shown in fig. 3.
As shown in fig. 3, for target user xiThe set of relational users whose correspondence to the meta paths 1 to P is obtained in the manner as described aboveAnd sets of relationshipsEach set of relational users comprises a plurality of second user characteristics and each set of relations comprises a plurality of relational characteristics. Therefore, as shown in FIG. 2, first, for each meta-path, each node in the meta-path, that is, each of a plurality of user nodes included in the meta-path, is identifiedAnd splicing the second characteristic features and the relational characteristics of the users of the nodes to obtain a plurality of relational user characteristics.
For example, for the meta path { user a, user C, user D }, which includes three nodes, user a, user C, and user D, the second individual features and the relationship features of user a, user C, and user D are respectively spliced to obtain three relationship user features, for example, denoted as h1,h2And h3。
Next, the first individual feature of the target user is spliced with the plurality of relational user features to obtain a plurality of first splicing features. For example, the first individual feature is denoted as htargetThen, respectively combine it with h1,h2And h3Splicing to obtain three first splicing characteristics htarget,h1],[htarget,h2]And [ h ]target,h3]。
Then, the at least one attention mechanism is used to obtain a plurality of sub-attention vectors corresponding to each meta-path based on the plurality of first stitching features, respectively, and the plurality of sub-attention vectors are concatenated to obtain the attention feature vector.
Therefore, in the user classification method according to the embodiment of the present application, using at least one attention mechanism based on the first individual feature, the second individual feature and the relationship feature of the concatenation to obtain an attention feature vector includes: for each meta path, splicing the second characteristic and the relation characteristic of each relation user in the plurality of relation users to obtain a plurality of relation user characteristics, wherein the relation user characteristics are the spliced second characteristic and the relation characteristic; splicing the plurality of relational user features with the first individual feature to obtain a plurality of first splicing features, wherein the first splicing features are the spliced first individual feature, second individual feature and splicing feature; using at least one attention mechanism based on the first splicing features to obtain a plurality of sub-attention vectors corresponding to each meta-path, wherein the sub-attention vectors are feature vectors used for representing second feature of the corresponding relation user and importance degree of the relation feature relative to the target user corresponding to each meta-path; and concatenating the plurality of sub-attention vectors to obtain the attention feature vector.
It should be noted that, in the embodiment of the present application, the first individual feature of the target user and the respective second individual feature and relationship feature of each relationship user may have different sizes, and therefore, before the features are spliced, the feature vectors having the same size need to be obtained by transforming matrix mapping. For example, a fully-connected layer may be used to convert different sized features into the same sized feature vector.
Therefore, in the user classification method according to the embodiment of the present application, for each meta path, the concatenating the second characteristic feature and the relationship characteristic feature of each user of the plurality of users to obtain a plurality of relationship user characteristics includes: for each meta path, mapping the respective second individual characteristic and relationship characteristic of each user in the plurality of users into a second individual characteristic vector and relationship characteristic vector with the same length through a transformation matrix, wherein the transformation matrix is used for converting different vectors into the same length; and splicing the second individual feature vector and the relation feature vector with the same length into the plurality of relation user features.
Moreover, in the user classification method according to the embodiment of the present application, the stitching the plurality of relationship user features and the first individual feature to obtain a plurality of first stitched features includes: mapping the first individual feature into a first individual feature vector having the same size as the plurality of relational user features through a transformation matrix; and splicing the plurality of relational user features with the same size and the first individual feature into the plurality of first splicing features.
Still taking the above example as an example, for the three first stitching features h obtainedtarget,h1],[htarget,h2]And [ h ]target,h3]The attention vector is obtained by an attention mechanism, which is expressed, for example, as:
h′i=[htar9et,hi]
zi=tanh(Whi)
where W and q are parameters to be trained in the attention mechanism and L represents the number of objects to be computed in different attention mechanisms. For example, if different users in the set of users for each relationship are to be computed by the node attention mechanism versus target user xiL is the number of nodes. If different features of the user set for each relationship are to be calculated by the feature attention mechanism for target user xiL is the number of features. And if the semantic attention mechanism is to be passed, calculating different relations to the target user xiL is the number of relationship types, i.e. the number of categories of meta-paths.
In the exemplary network structure shown in fig. 3, the attention feature vector is obtained using three of the node attention mechanism, the feature attention mechanism, and the semantic attention mechanism, but as described above, in the user classification method according to an embodiment of the present application, the attention feature vector may also be obtained using any one or more two attention mechanisms of the node attention mechanism, the feature attention mechanism, and the semantic attention mechanism based on the attention mechanism as described above.
With continued reference to FIG. 3, the different user-to-target user x in the user set for each relationship is first computed by the node attention mechanismiThat is, according to the attention mechanism described above, a node attention vector is obtained based on the plurality of first stitched features, and then the node attention vector is multiplied by each of the relationship user features to obtain a plurality of node features. Or, alsoThe node attention vector may be multiplied by the each of the relational user features based on a different weight to obtain a plurality of node features. Next, the sub-attention vector is obtained based on the sum of the plurality of node features, that is, if only the node attention mechanism is adopted, the plurality of node features may be directly summed to obtain the sub-attention vector, or based on the sum of the plurality of node features, the feature attention mechanism and/or the semantic attention mechanism may be further used to obtain the sub-attention vector.
Therefore, in the user classification method according to an embodiment of the present application, using at least one attention mechanism based on the plurality of stitched features to obtain a plurality of sub-attention vectors corresponding to each meta-path includes: using a node attention mechanism based on the plurality of first stitching features to obtain a node attention vector, the node attention mechanism being used to calculate degrees of importance of different users as nodes to the target user, and the node attention vector being a feature vector in which features of different users have different weights based on the degrees of importance; obtaining a plurality of node features from the node attention vector and the each relational user feature; and obtaining the sub-attention vector based on the sum of the plurality of node features.
With continued reference to FIG. 3, with continued use of the feature attention mechanism, the plurality of node features are first summed for each meta-path to obtain a sum feature vector. Then, according to the attention mechanism as described above, the sum feature vector is first stitched with the first individual feature to obtain a second stitched vector, then the feature attention mechanism is used based on the second stitched vector to obtain a feature attention vector, and then the feature attention vector is multiplied with the second stitched vector to obtain a user set feature vector. Alternatively, the feature attention vector and the second stitching vector may be multiplied based on different weights to obtain a user set feature vector. Here, the user set feature vector represents a user set corresponding to each meta path, for example, a feature vector of a user set { user a, user C, user D } as described above.
Similarly, if the semantic attention mechanism is not used, the user set feature vector corresponding to each meta path can be directly used as the sub-attention vector. Alternatively, a semantic attention mechanism may be further used to obtain the sub-attention vector.
Therefore, in the above user classification method, obtaining the sub-attention vector based on the sum of the plurality of node features includes: for each meta-path, obtaining a sum feature vector from the plurality of node features; stitching the sum feature vector with the first individual feature to obtain a second stitched vector; using a feature attention mechanism based on the second stitching vector to obtain a feature attention vector, the feature attention mechanism for calculating degrees of importance of different types of features to the target user, and the feature attention vector being a feature vector in which different types of features have different weights based on the degrees of importance; obtaining a user set feature vector from the feature attention vector and the second stitching vector; and obtaining the sub-attention vector based on a plurality of user set feature vectors corresponding to a plurality of meta-paths.
With continued reference to fig. 3, in a case where the semantic attention mechanism is continuously used, according to the attention mechanism as described above, the plurality of user set feature vectors are first spliced with the first individual feature to obtain a plurality of third spliced vectors, then the semantic attention mechanism is used based on the third spliced vectors to obtain a plurality of semantic attention vectors, and finally each corresponding feature-level feature vector is multiplied with the semantic attention vector to obtain the sub-attention vector. Alternatively, each corresponding feature-level feature vector may be multiplied by a semantic attention vector with a different weight to obtain the sub-attention vector.
That is, in the above-described user classification method, obtaining the sub-attention vector based on a plurality of user set feature vectors corresponding to a plurality of meta paths includes: stitching the plurality of user set feature vectors with the first individual feature respectively to obtain a plurality of third stitching vectors; using a semantic attention mechanism based on the third stitching vector to obtain a plurality of semantic attention vectors, the semantic attention mechanism for calculating degrees of importance of different relationship types to the target user, and the feature attention vectors being feature vectors in which features of different relationship types have different weights based on the degrees of importance; and obtaining the sub-attention vector from each corresponding user set feature vector and semantic attention vector.
In this way, through the multi-attention mechanism as shown in fig. 3, multi-sense-field capture of individual and relationship importance levels can be achieved, thereby obtaining accurate classification results.
Exemplary method two
FIG. 4 illustrates a flow chart of a financial user overdue identification method according to an embodiment of the application.
As shown in fig. 4, the method for identifying the overdue financial users according to the embodiment of the present application includes: s210, acquiring a first individual feature of a financial user, wherein the first individual feature is used for representing an individual attribute of the financial user; s220, traversing by using a meta-path method aiming at different relation types of relation users of the financial users to acquire a plurality of meta-paths each including a plurality of relation users, wherein the meta-path method is used for acquiring the plurality of relation users similar based on different relation types in the relation users through similarity search; s230, acquiring a second individual characteristic and a relation characteristic of the relation user in each meta-path, wherein the second individual characteristic is used for representing the individual attribute of the relation user, and the relation characteristic is used for representing the relation between the relation user and the financial user; s240, using at least one attention mechanism based on the first individual feature, the second individual feature and the relation feature of the concatenation to obtain an attention feature vector, wherein the at least one attention mechanism is used for calculating importance degrees of different features to the target user, and the attention feature vector is a feature vector in which different features have different weights based on the importance degrees; and S250, identifying whether the financial user is overdue or not based on the attention feature vector by using a multi-layer sensor network, wherein the multi-layer sensor network is used for classifying based on the feature vector and is provided with a label for representing the overdue of the financial user.
Here, in the financial user overdue identification method according to the embodiment of the present application as shown in fig. 4, the target user is a user who needs to identify whether the financial is overdue, and the related user may be a user having a financial relationship with the target user, such as a user having a financial transaction flow.
Moreover, by the financial user overdue identification method according to the embodiment of the present application as shown in fig. 4, the result of whether the user is overdue or not may be obtained, so that the result of whether the user is overdue or not may be provided to a financial institution such as a bank, particularly, for example, a loan department thereof, thereby promoting the loan security of the loan department.
Those skilled in the art can understand that other details of the financial user overdue identification method according to the embodiment of the present application are completely the same as the corresponding details of the user classification method according to the embodiment of the present application described with reference to fig. 1 to 3, and are not described again to avoid redundancy.
Exemplary method three
Fig. 5 illustrates a flowchart of a search recommendation method of a target user according to an embodiment of the present application.
As shown in fig. 5, a search recommendation method for a target user according to an embodiment of the present application includes: s310, acquiring a first individual feature of a target user, wherein the first individual feature is used for representing an individual attribute of the target user; s320, traversing by using a meta-path method aiming at different relation types of the relation users of the target user to obtain a plurality of meta-paths each containing a plurality of relation users, wherein the meta-path method is used for obtaining the plurality of relation users based on similarity of different relation types in the relation users through similarity search; s330, acquiring a second individual characteristic and a relation characteristic of the relation user in each meta-path, wherein the second individual characteristic is used for representing the individual attribute of the relation user, and the relation characteristic is used for representing the relation between the relation user and the target user; s340, using at least one attention mechanism based on the first individual feature, the second individual feature and the relation feature of the concatenation to obtain an attention feature vector, wherein the at least one attention mechanism is used for calculating importance degrees of different features to the target user, and the attention feature vector is a feature vector in which different features have different weights based on the importance degrees; s350, classifying the target user based on the attention feature vector by using a multilayer perceptron network, wherein the multilayer perceptron network is used for classifying based on the feature vector and is provided with a label for expressing a category recommended to the target user for searching; and S360, searching and recommending to the target user according to the category.
Here, in the search recommendation method for a target user according to the embodiment of the present application as shown in fig. 5, the target user is a user who needs to make a search recommendation, such as a user who is browsing a shopping website, and the related user may be a user who has a social relationship with the target user, or another user who has purchased the same item on the shopping website with the target user, and so on.
In addition, by the search recommendation method for the target user according to the embodiment of the application as shown in fig. 5, the search recommendation result of the target user can be obtained, so that the search recommendation result of the target user can be provided to other users such as a shopping website and the like who need the search recommendation result of the user, and the other users can provide the search recommendation result conforming to the target user.
Those skilled in the art can understand that other details of the search recommendation method for the target user according to the embodiment of the present application are completely the same as the corresponding details in the user classification method according to the embodiment of the present application described previously with reference to fig. 1 to 3, and are not described again to avoid redundancy.
Exemplary method four
Fig. 6 illustrates a flow chart of a false transaction identification method according to an embodiment of the application.
As shown in fig. 6, a false transaction identification method according to an embodiment of the present application includes: s410, acquiring a first individual characteristic of a transaction user, wherein the first individual characteristic is used for representing an individual attribute of the financial user; s420, traversing by using a meta-path method aiming at different relation types of relation users of the transaction user to obtain a plurality of meta-paths each containing a plurality of relation users, wherein the meta-path method is used for obtaining the plurality of relation users based on similarity of different relation types in the relation users through similarity search; s430, acquiring a second individual characteristic and a relation characteristic of the relation user in each meta-path, wherein the second individual characteristic is used for representing individual attributes of the relation user, and the relation characteristic is used for representing the relation between the relation user and the transaction user; s440, using at least one attention mechanism based on the first individual feature, the second individual feature and the relationship feature of the concatenation to obtain an attention feature vector, the at least one attention mechanism being used for calculating degrees of importance of different features to the target user, and the attention feature vector being a feature vector in which different features have different weights based on the degrees of importance; and S450, identifying whether the transaction user carries out false transaction or not based on the attention feature vector by using a multi-layer perceptron network, wherein the multi-layer perceptron network is used for carrying out classification based on the feature vector and is provided with a label used for representing false transaction.
Here, in the false transaction identification method according to the embodiment of the present application as shown in fig. 6, the target user is a user who needs to identify whether a false transaction is caused, such as a transaction user of a specific website, and the relationship user may be a user who has a transaction relationship with the target user, such as a user who has a corresponding trade relationship as a buyer or a seller, and so on.
In addition, by the false transaction identification method according to the embodiment of the present application as shown in fig. 6, a classification result of whether the target user is a false transaction can be obtained, so that a determination result of whether the target user is a false transaction can be provided to a service provider of a website, and the service provider can accurately identify a user who has performed a false transaction.
Those skilled in the art can understand that other details of the false transaction identification method according to the embodiment of the present application are exactly the same as corresponding details in the user classification method according to the embodiment of the present application described previously with reference to fig. 1 to 3, and are not described again to avoid redundancy.
Exemplary method five
FIG. 7 illustrates a flow chart of a method of identifying a user's intent to pick up a coupon according to an embodiment of the application.
As shown in fig. 7, the method for identifying the user's intention to receive a coupon according to an embodiment of the present application includes: s510, acquiring a first individual characteristic of a user who receives a coupon, wherein the first individual characteristic is used for representing an individual attribute of the user; s520, traversing by using a meta-path method aiming at different relation types of relation users of the users to acquire a plurality of meta-paths each including a plurality of relation users, wherein the meta-path method is used for acquiring the plurality of relation users similar based on different relation types in the relation users through similarity search; s530, obtaining a second individual characteristic and a relation characteristic of the relation user in each meta-path, wherein the second individual characteristic is used for representing the individual attribute of the relation user, and the relation characteristic is used for representing the relation between the relation user and the user receiving the discount coupons; s540, using at least one attention mechanism based on the first individual feature, the second individual feature and the relationship feature of the concatenation to obtain an attention feature vector, wherein the at least one attention mechanism is used for calculating importance degrees of different features to the target user, and the attention feature vector is a feature vector in which different features have different weights based on the importance degrees; and S550, identifying whether the coupon picked up by the user is malicious intention or not based on the attention feature vector by using a multi-layer perceptron network, wherein the multi-layer perceptron network is used for classifying based on the feature vector and is provided with a label used for representing the malicious intention of the user.
Here, in the method for identifying the user's intention to receive the coupon according to the embodiment of the present application as shown in fig. 7, the target user is a user who has received or is to receive the coupon, such as a user of a shopping site, and the relationship user may be a user who has a social relationship with the target user, such as a relationship user who forwards coupon information through a social network site by the target user, or the like.
Moreover, by the method for identifying the user's intention to receive the coupon according to the embodiment of the present application as shown in fig. 7, the classification result of whether the target user receives the coupon is a malicious intention can be obtained, so that the provider of the coupon can provide the determination result of whether the target user receives the coupon is the malicious intention, thereby enabling the provider of the coupon to accurately identify the user who maliciously receives the coupon, and reducing the loss of the provider of the coupon.
Further, in the method of recognizing the user's intention to receive the coupon according to the embodiment of the present application as shown in fig. 7, the coupon may be a full discount coupon at the time of shopping or may be in the form of a simple cash red pack. The provider of the coupon is not limited to the shopping site or the seller thereof, but includes other users who provide the cash red package.
Those skilled in the art can understand that other details of the method for identifying the user's intention to receive the coupon according to the embodiment of the present application are completely the same as the corresponding details in the user classification method according to the embodiment of the present application described previously with reference to fig. 1 to 3, and are not described again to avoid redundancy.
Illustrative apparatus one
Fig. 8 illustrates a block diagram of a user classification device according to an embodiment of the application.
As shown in fig. 8, a user classification apparatus 600 according to an embodiment of the present application includes: a target feature obtaining unit 610, configured to obtain a first individual feature of a target user, where the first individual feature is used to represent an individual attribute of the target user; a meta-path traversing unit 620, configured to traverse, using a meta-path method for different relationship types of the relationship user of the target user, to obtain multiple meta-paths each including multiple relationship users, where the meta-path method is used to obtain, through similarity search, the multiple relationship users that are similar based on different relationship types among the relationship users; a relationship characteristic obtaining unit 630, configured to obtain a second individual characteristic and a relationship characteristic of the relationship user in each meta path, where the second individual characteristic is used to represent an individual attribute of the relationship user, and the relationship characteristic is used to represent a relationship between the relationship user and the target user; an attention mechanism unit 640, configured to use at least one attention mechanism based on the first individual feature, the second individual feature and the relationship feature of the stitching to obtain an attention feature vector, wherein the at least one attention mechanism is used for calculating importance degrees of different features to the target user, and the attention feature vector is a feature vector in which different features have different weights based on the importance degrees; and a classification result obtaining unit 650 configured to obtain a classification result of the target user based on the attention feature vector using a multi-layered perceptron network, the multi-layered perceptron network configured to perform classification based on the feature vector.
In one example, in the user classification apparatus 600 described above, the meta path has a corresponding associated relationship type.
In one example, in the user classification apparatus 600 described above, the attention mechanism unit 640 is configured to: for each meta path, splicing the second characteristic and the relation characteristic of each relation user in the plurality of relation users to obtain a plurality of relation user characteristics, wherein the relation user characteristics are the spliced second characteristic and the relation characteristic; splicing the plurality of relational user features with the first individual feature to obtain a plurality of first splicing features, wherein the first splicing features are the spliced first individual feature, second individual feature and splicing feature; using at least one attention mechanism based on the first splicing features to obtain a plurality of sub-attention vectors corresponding to each meta-path, wherein the sub-attention vectors are feature vectors used for representing second feature of the corresponding relation user and importance degree of the relation feature relative to the target user corresponding to each meta-path; and concatenating the plurality of sub-attention vectors to obtain the attention feature vector.
In one example, in the user classification apparatus 600, the attention mechanism unit 640 concatenates the second characteristic feature and the relationship feature of each user of the plurality of users to obtain a plurality of relationship user characteristics for each meta path includes: for each meta path, mapping the respective second individual characteristic and relationship characteristic of each user in the plurality of users into a second individual characteristic vector and relationship characteristic vector with the same length through a transformation matrix, wherein the transformation matrix is used for converting different vectors into the same length; and splicing the second individual feature vector and the relation feature vector with the same length into the plurality of relation user features.
In one example, in the user classification apparatus 600 described above, the using at least one attention mechanism by the attention mechanism unit 640 to obtain a plurality of sub-attention vectors corresponding to each meta-path based on the plurality of stitching features includes: using a node attention mechanism based on the plurality of first stitching features to obtain a node attention vector, the node attention mechanism being used to calculate degrees of importance of different users as nodes to the target user, and the node attention vector being a feature vector in which features of different users have different weights based on the degrees of importance; obtaining a plurality of node features from the node attention vector and the each relational user feature; and obtaining the sub-attention vector based on the sum of the plurality of node features.
In one example, in the user classification apparatus 600 described above, the obtaining of the sub-attention vector by the attention mechanism unit 640 based on the sum of the plurality of node features includes: for each meta-path, obtaining a sum feature vector from the plurality of node features; stitching the sum feature vector with the first individual feature to obtain a second stitched vector; using a feature attention mechanism based on the second stitching vector to obtain a feature attention vector, the feature attention mechanism for calculating degrees of importance of different types of features to the target user, and the feature attention vector being a feature vector in which different types of features have different weights based on the degrees of importance; obtaining a user set feature vector from the feature attention vector and the second stitching vector; and obtaining the sub-attention vector based on a plurality of user set feature vectors corresponding to a plurality of meta-paths.
In one example, in the user classification apparatus 600 described above, the obtaining of the sub-attention vector by the attention mechanism unit 640 based on a plurality of user set feature vectors corresponding to a plurality of meta paths includes: stitching the plurality of user set feature vectors with the first individual feature respectively to obtain a plurality of third stitching vectors; using a semantic attention mechanism based on the third stitching vector to obtain a plurality of semantic attention vectors, the semantic attention mechanism for calculating degrees of importance of different relationship types to the target user, and the feature attention vectors being feature vectors in which features of different relationship types have different weights based on the degrees of importance; and obtaining the sub-attention vector from each corresponding user set feature vector and semantic attention vector.
In one example, in the user classifying apparatus 600, the target user includes one of a user account, a user device, and a user item.
In one example, in the user classification apparatus 600, the different relationship types include one of a social relationship, a fund relationship, and a device login relationship.
In one example, in the user classifying means 600, the classification result includes one of a financial overdue result, a search recommendation result, and a false transaction result.
Here, it will be understood by those skilled in the art that the detailed functions and operations of the respective units and modules in the user classifying apparatus 600 have been described in detail in the above description of the user classifying method with reference to fig. 1 to 3, and thus, a repetitive description thereof will be omitted.
As described above, the user classifying apparatus 600 according to the embodiment of the present application may be implemented in various terminal devices, such as a server of a financial anti-fraud website. In one example, the user classifying apparatus 600 according to the embodiment of the present application may be integrated into a terminal device as a software module and/or a hardware module. For example, the user classifying means 600 may be a software module in the operating system of the terminal device, or may be an application developed for the terminal device; of course, the user classifying means 600 may also be one of many hardware modules of the terminal device.
Alternatively, in another example, the user classifying means 600 and the terminal device may be separate devices, and the user classifying means 600 may be connected to the terminal device through a wired and/or wireless network and transmit the interactive information according to an agreed data format.
Exemplary apparatus two
Fig. 9 illustrates a block diagram of a financial user overdue identification apparatus according to an embodiment of the present application.
As shown in fig. 9, the financial user overdue identification apparatus 700 according to the embodiment of the present application includes: a financial user characteristic acquiring unit 710, configured to acquire a first individual characteristic of a financial user, where the first individual characteristic is used to represent an individual attribute of the financial user; a financial element path traversing unit 720, configured to traverse, by using an element path method, for different relationship types of relationship users of the financial user to obtain multiple element paths each including multiple relationship users, where the element path method is used to obtain, through similarity search, the multiple relationship users similar based on different relationship types among the relationship users; a financial relationship characteristic obtaining unit 730, configured to obtain a second individual characteristic and a relationship characteristic of the relationship user in each meta path, where the second individual characteristic is used to represent an individual attribute of the relationship user, and the relationship characteristic is used to represent a relationship between the relationship user and the financial user; a financial attention mechanism unit 740, configured to use at least one attention mechanism based on the first individual feature, the second individual feature and the relationship feature of the concatenation to obtain an attention feature vector, wherein the at least one attention mechanism is used to calculate importance degrees of different features to the target user, and the attention feature vector is a feature vector in which different features have different weights based on the importance degrees; and a financial user overdue identification unit 750 configured to identify whether the financial user is overdue based on the attention feature vector using a multi-layered sensor network, the multi-layered sensor network being configured to classify based on the feature vector and being provided with a tag indicating that the financial user is overdue.
Here, it can be understood by those skilled in the art that the specific functions and operations of the respective units and modules in the above-described financial user overdue identification apparatus 700 exactly correspond to those of the respective units and modules in the user classifying apparatus 600 according to the embodiment of the present application as shown in fig. 8, and thus, the repetitive description thereof will be omitted.
Exemplary apparatus III
Fig. 10 illustrates a block diagram of a search recommendation apparatus of a target user according to an embodiment of the present application.
As shown in fig. 10, the search recommendation apparatus 800 of the target user according to the embodiment of the present application includes: a search user characteristic obtaining unit 810, configured to obtain a first individual characteristic of a target user, where the first individual characteristic is used to represent an individual attribute of the target user; a search meta-path traversing unit 820, configured to traverse, using a meta-path method for different relationship types of the relationship user of the target user, to obtain multiple meta-paths each including multiple relationship users, where the meta-path method is used to obtain, through similarity search, the multiple relationship users similar based on different relationship types in the relationship users; a search relationship feature obtaining unit 830, configured to obtain a second feature and a relationship feature of the relationship user in each meta path, where the second feature is used to represent an individual attribute of the relationship user, and the relationship feature is used to represent a relationship between the relationship user and the target user; a search attention mechanism unit 840 for using at least one attention mechanism based on the first individual feature, the second individual feature and the relationship feature of the concatenation to obtain an attention feature vector, the at least one attention mechanism being used to calculate degrees of importance of different features to the target user, and the attention feature vector being a feature vector in which different features have different weights based on the degrees of importance; a user classifying unit 850 configured to classify the target user based on the attention feature vector using a multi-layer perceptron network, the multi-layer perceptron network being configured to classify based on the feature vector and being provided with a tag indicating a category for making a search recommendation to the target user; and a search recommendation unit 860, configured to perform search recommendation to the target user according to the category.
Here, it can be understood by those skilled in the art that the specific functions and operations of the respective units and modules in the search recommending apparatus 800 of the target user described above completely correspond to those of the respective units and modules in the user classifying apparatus 600 according to the embodiment of the present application as shown in fig. 8, and thus, the repeated description thereof will be omitted.
Exemplary apparatus four
FIG. 11 illustrates a block diagram of a false transaction identification device according to an embodiment of the present application.
As shown in fig. 11, the false transaction identification device 900 according to the embodiment of the present application includes: a trading user characteristic obtaining unit 910, configured to obtain a first individual characteristic of a trading user, where the first individual characteristic is used to represent an individual attribute of the financial user; a transaction meta-path traversing unit 920, configured to traverse, by using a meta-path method, for different relationship types of relationship users of the transaction user to obtain multiple meta-paths each including multiple relationship users, where the meta-path method is used to obtain, through similarity search, the multiple relationship users similar based on the different relationship types in the relationship users; a transaction relationship characteristic obtaining unit 930, configured to obtain a second individual characteristic and a relationship characteristic of the relationship user in each meta path, where the second individual characteristic is used to represent an individual attribute of the relationship user, and the relationship characteristic is used to represent a relationship between the relationship user and the transaction user; a transaction attention mechanism unit 940 for using at least one attention mechanism based on the stitched first individual feature, the second individual feature and the relationship feature to obtain an attention feature vector, wherein the at least one attention mechanism is used for calculating importance degrees of different features to the target user, and the attention feature vector is a feature vector in which different features have different weights based on the importance degrees; and a false transaction identification unit 950 for identifying whether the transaction user has performed a false transaction based on the attention feature vector using a multi-layered sensor network for classifying based on the feature vector and provided with a tag for representing a false transaction.
Here, it will be understood by those skilled in the art that the specific functions and operations of the respective units and modules in the above-described false transaction recognition apparatus 900 exactly correspond to those of the respective units and modules in the user classifying apparatus 600 according to the embodiment of the present application as shown in fig. 8, and thus, the repetitive description thereof will be omitted.
Schematic device five
Fig. 12 illustrates a block diagram of an apparatus for recognizing a user's intention to take a coupon according to an embodiment of the present application.
As shown in fig. 12, an apparatus 1000 for recognizing a user's intention to get a coupon according to an embodiment of the present application includes: an intended user characteristic obtaining unit 1010 for obtaining a first individual characteristic of a user who picks up a coupon, the first individual characteristic being indicative of an individual attribute of the user; an ideogram path traversing unit 1020, configured to traverse, using a meta path method, for different relationship types of the relationship users of the user, to obtain multiple meta paths each including multiple relationship users, where the meta path method is used to obtain, through similarity search, the multiple relationship users similar based on different relationship types among the relationship users; an intention relation feature obtaining unit 1030, configured to obtain a second individual feature and a relation feature of the relation user in each meta path, where the second individual feature is used to represent an individual attribute of the relation user, and the relation feature is used to represent a relation between the relation user and the user who receives the coupon; an intention attention mechanism unit 1040 for using at least one attention mechanism based on the first individual feature, the second individual feature and the relationship feature of the stitching to obtain an attention feature vector, the at least one attention mechanism being used for calculating degrees of importance of different features to the target user, and the attention feature vector being a feature vector in which different features have different weights based on the degrees of importance; and a malicious intent recognition unit 1050 configured to recognize whether the coupon picked up by the user is a malicious intent based on the attention feature vector using a multi-layer perceptron network, the multi-layer perceptron network being configured to classify based on the feature vector and being provided with a tag representing the malicious intent of the user.
Here, it can be understood by those skilled in the art that the specific functions and operations of the respective units and modules in the above-described recognition apparatus 1000 of the user's intention to take a coupon fully correspond to those of the respective units and modules in the user classifying apparatus 600 according to the embodiment of the present application as shown in fig. 8, and thus, a repetitive description thereof will be omitted.
Exemplary electronic device
Next, an electronic apparatus according to an embodiment of the present application is described with reference to fig. 13.
FIG. 13 illustrates a block diagram of an electronic device in accordance with an embodiment of the present application.
As shown in fig. 13, the electronic device 10 includes one or more processors 11 and a memory 12.
The processor 13 may be a Central Processing Unit (CPU) or other form of processing unit having data processing capabilities and/or instruction execution capabilities, and may control other components in the electronic device 10 to perform desired functions.
In one example, the electronic device 10 may further include: an input device 13 and an output device 14, which are interconnected by a bus system and/or other form of connection mechanism (not shown).
The input device 13 may include, for example, a keyboard, a mouse, and the like.
The output device 14 can output various information including the user classification result to the outside. The output devices 14 may include, for example, a display, speakers, a printer, and a communication network and its connected remote output devices, among others.
Of course, for simplicity, only some of the components of the electronic device 10 relevant to the present application are shown in fig. 13, and components such as buses, input/output interfaces, and the like are omitted. In addition, the electronic device 10 may include any other suitable components depending on the particular application.
Exemplary computer program product and computer-readable storage Medium
In addition to the above methods and apparatus, embodiments of the present application may also be a computer program product comprising computer program instructions that, when executed by a processor, cause the processor to perform the steps of the user classification method, the financial user overdue identification method, the search recommendation method for a target user, the fake transaction identification method, and the identification method of the user's intention to receive a coupon according to various embodiments of the present application described in the "exemplary methods" section above in this specification.
The computer program product may be written with program code for performing the operations of embodiments of the present application in any combination of one or more programming languages, including an object oriented programming language such as Java, C + + or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code may execute entirely on the first user computing device, partly on the first user device, as a stand-alone software package, partly on the first user computing device and partly on a remote computing device, or entirely on the remote computing device or server.
Furthermore, embodiments of the present application may also be a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, cause the processor to perform the steps of the user classification method, the financial user overdue identification method, the search recommendation method of the target user, the false transaction identification method, and the identification method of the user's intention to receive the coupon according to various embodiments of the present application described in the above-mentioned "exemplary methods" section of this specification.
The computer-readable storage medium may take any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may include, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or a combination of any of the foregoing. More specific examples (a non-exhaustive list) of the readable storage medium include: an electrical connection having one or more wires, a portable disk, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.
The foregoing describes the general principles of the present application in conjunction with specific embodiments, however, it is noted that the advantages, effects, etc. mentioned in the present application are merely examples and are not limiting, and they should not be considered essential to the various embodiments of the present application. Furthermore, the foregoing disclosure of specific details is for the purpose of illustration and description and is not intended to be limiting, since the foregoing disclosure is not intended to be exhaustive or to limit the disclosure to the precise details disclosed.
The block diagrams of devices, apparatuses, systems referred to in this application are only given as illustrative examples and are not intended to require or imply that the connections, arrangements, configurations, etc. must be made in the manner shown in the block diagrams. These devices, apparatuses, devices, systems may be connected, arranged, configured in any manner, as will be appreciated by those skilled in the art. Words such as "including," "comprising," "having," and the like are open-ended words that mean "including, but not limited to," and are used interchangeably therewith. The words "or" and "as used herein mean, and are used interchangeably with, the word" and/or, "unless the context clearly dictates otherwise. The word "such as" is used herein to mean, and is used interchangeably with, the phrase "such as but not limited to".
It should also be noted that in the devices, apparatuses, and methods of the present application, the components or steps may be decomposed and/or recombined. These decompositions and/or recombinations are to be considered as equivalents of the present application.
The previous description of the disclosed aspects is provided to enable any person skilled in the art to make or use the present application. Various modifications to these aspects will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other aspects without departing from the scope of the application. Thus, the present application is not intended to be limited to the aspects shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
The foregoing description has been presented for purposes of illustration and description. Furthermore, the description is not intended to limit embodiments of the application to the form disclosed herein. While a number of example aspects and embodiments have been discussed above, those of skill in the art will recognize certain variations, modifications, alterations, additions and sub-combinations thereof.
Claims (21)
1. A method for classifying a user, comprising:
acquiring a first individual feature of a target user, wherein the first individual feature is used for representing an individual attribute of the target user;
traversing by using a meta-path method aiming at different relation types of the relation users of the target user to obtain a plurality of meta-paths each containing a plurality of relation users, wherein the meta-path method is used for obtaining the plurality of relation users based on similarity of different relation types in the relation users through similarity search;
acquiring a second individual characteristic and a relation characteristic of a relation user in each meta path, wherein the second individual characteristic is used for representing individual attributes of the relation user, and the relation characteristic is used for representing the relation between the relation user and the target user;
using at least one attention mechanism for calculating degrees of importance of different features to the target user based on the stitched first individual feature, the second individual feature and the relationship feature to obtain an attention feature vector, and the attention feature vector is a feature vector in which different features have different weights based on the degrees of importance; and
obtaining a classification result of the target user based on the attention feature vector using a multi-layer perceptron network for classifying based on feature vectors.
2. The user classification method according to claim 1, characterized in that the meta path has a corresponding associated relationship type.
3. The method of claim 2, wherein using at least one attention mechanism based on the first individual feature, the second individual feature and the relationship feature of the concatenation to obtain an attention feature vector comprises:
for each meta path, splicing the second characteristic and the relation characteristic of each relation user in the plurality of relation users to obtain a plurality of relation user characteristics, wherein the relation user characteristics are the spliced second characteristic and the relation characteristic;
splicing the plurality of relational user features with the first individual feature to obtain a plurality of first splicing features, wherein the first splicing features are the spliced first individual feature, second individual feature and splicing feature;
using at least one attention mechanism based on the first splicing features to obtain a plurality of sub-attention vectors corresponding to each meta-path, wherein the sub-attention vectors are feature vectors used for representing second feature of the corresponding relation user and importance degree of the relation feature relative to the target user corresponding to each meta-path; and
concatenating the plurality of sub-attention vectors to obtain the attention feature vector.
4. The method of claim 3, wherein for each meta-path, concatenating the second individual trait and the relational trait of each of the plurality of users to obtain a plurality of relational user traits comprises:
for each meta path, mapping the respective second individual characteristic and relationship characteristic of each user in the plurality of users into a second individual characteristic vector and relationship characteristic vector with the same length through a transformation matrix, wherein the transformation matrix is used for converting different vectors into the same length; and
and splicing the second individual feature vector and the relation feature vector with the same length into the plurality of relation user features.
5. The user classification method according to claim 3, wherein using at least one attention mechanism based on the plurality of stitched features to obtain a plurality of sub-attention vectors corresponding to each meta-path comprises:
using a node attention mechanism based on the plurality of first stitching features to obtain a node attention vector, the node attention mechanism being used to calculate degrees of importance of different users as nodes to the target user, and the node attention vector being a feature vector in which features of different users have different weights based on the degrees of importance;
obtaining a plurality of node features from the node attention vector and the each relational user feature; and
obtaining the sub-attention vector based on a sum of the plurality of node features.
6. The user classification method according to claim 5, wherein obtaining the sub-attention vector based on the sum of the plurality of node features comprises:
for each meta-path, obtaining a sum feature vector from the plurality of node features;
stitching the sum feature vector with the first individual feature to obtain a second stitched vector;
using a feature attention mechanism based on the second stitching vector to obtain a feature attention vector, the feature attention mechanism for calculating degrees of importance of different types of features to the target user, and the feature attention vector being a feature vector in which different types of features have different weights based on the degrees of importance;
obtaining a user set feature vector from the feature attention vector and the second stitching vector; and
the sub-attention vector is obtained based on a plurality of user set feature vectors corresponding to a plurality of meta-paths.
7. The user classification method according to claim 6, wherein obtaining the sub-attention vector based on a plurality of user set feature vectors corresponding to a plurality of meta-paths comprises:
stitching the plurality of user set feature vectors with the first individual feature respectively to obtain a plurality of third stitching vectors;
using a semantic attention mechanism based on the third stitching vector to obtain a plurality of semantic attention vectors, the semantic attention mechanism for calculating degrees of importance of different relationship types to the target user, and the feature attention vectors being feature vectors in which features of different relationship types have different weights based on the degrees of importance; and
the sub-attention vector is obtained from each corresponding user set feature vector and semantic attention vector.
8. The user classification method according to any one of claims 1 to 7, wherein the target user comprises one of a user account, a user device, and a user item.
9. The user classification method according to claim 8, wherein the different relationship types include one of a social relationship, a funding relationship, and a device login relationship.
10. The user categorization method of claim 9 wherein the categorization result comprises one of a financial overdue result, a search recommendation result, a false deal result.
11. A financial user overdue identification method is characterized by comprising the following steps:
acquiring a first individual feature of a financial user, wherein the first individual feature is used for representing an individual attribute of the financial user;
traversing by using a meta-path method aiming at different relation types of relation users of the financial users to acquire a plurality of meta-paths each including a plurality of relation users, wherein the meta-path method is used for acquiring the plurality of relation users similar based on different relation types in the relation users through similarity search;
acquiring a second physical characteristic and a relation characteristic of the relation user in each meta-path, wherein the second physical characteristic is used for representing individual attributes of the relation user, and the relation characteristic is used for representing the relation between the relation user and the financial user;
using at least one attention mechanism for calculating degrees of importance of different features to the target user based on the stitched first individual feature, the second individual feature and the relationship feature to obtain an attention feature vector, and the attention feature vector is a feature vector in which different features have different weights based on the degrees of importance; and
identifying whether the financial user is overdue based on the attention feature vector by using a multi-layer sensor network, wherein the multi-layer sensor network is used for classifying based on the feature vector and is provided with a label used for representing the overdue of the financial user.
12. A search recommendation method for a target user is characterized by comprising the following steps:
acquiring a first individual feature of a target user, wherein the first individual feature is used for representing an individual attribute of the target user;
traversing by using a meta-path method aiming at different relation types of the relation users of the target user to obtain a plurality of meta-paths each containing a plurality of relation users, wherein the meta-path method is used for obtaining the plurality of relation users based on similarity of different relation types in the relation users through similarity search;
acquiring a second individual characteristic and a relation characteristic of a relation user in each meta path, wherein the second individual characteristic is used for representing individual attributes of the relation user, and the relation characteristic is used for representing the relation between the relation user and the target user;
using at least one attention mechanism for calculating degrees of importance of different features to the target user based on the stitched first individual feature, the second individual feature and the relationship feature to obtain an attention feature vector, and the attention feature vector is a feature vector in which different features have different weights based on the degrees of importance;
classifying the target user based on the attention feature vector by using a multi-layer perceptron network, wherein the multi-layer perceptron network is used for classifying based on the feature vector and is provided with a label for representing a category recommended to the target user for searching; and
and carrying out search recommendation to the target user according to the category.
13. A false transaction identification method, comprising:
acquiring a first individual characteristic of a transaction user, wherein the first individual characteristic is used for representing an individual attribute of the financial user;
traversing by using a meta-path method aiming at different relationship types of relationship users of the transaction user to obtain a plurality of meta-paths each containing a plurality of relationship users, wherein the meta-path method is used for obtaining the plurality of relationship users based on similarity of different relationship types in the relationship users through similarity search;
acquiring a second physical characteristic and a relation characteristic of a relation user in each meta-path, wherein the second physical characteristic is used for representing individual attributes of the relation user, and the relation characteristic is used for representing the relation between the relation user and the transaction user;
using at least one attention mechanism for calculating degrees of importance of different features to the target user based on the stitched first individual feature, the second individual feature and the relationship feature to obtain an attention feature vector, and the attention feature vector is a feature vector in which different features have different weights based on the degrees of importance; and
identifying whether the transaction user carries out false transactions based on the attention feature vector by using a multi-layer perceptron network, wherein the multi-layer perceptron network is used for classifying based on the feature vector and is provided with a label used for representing false transactions.
14. A method for identifying a user's intent to receive a coupon, comprising:
acquiring a first individual characteristic of a user receiving a coupon, wherein the first individual characteristic is used for representing an individual attribute of the user;
traversing by using a meta-path method aiming at different relation types of relation users of the users to acquire a plurality of meta-paths each including a plurality of relation users, wherein the meta-path method is used for acquiring the plurality of relation users similar based on different relation types in the relation users through similarity search;
acquiring a second individual characteristic and a relation characteristic of a relation user in each meta path, wherein the second individual characteristic is used for representing individual attributes of the relation user, and the relation characteristic is used for representing the relation between the relation user and a user receiving the coupons;
using at least one attention mechanism for calculating degrees of importance of different features to the target user based on the stitched first individual feature, the second individual feature and the relationship feature to obtain an attention feature vector, and the attention feature vector is a feature vector in which different features have different weights based on the degrees of importance; and
identifying whether the user is maliciously intentioned to pick up the coupon based on the attention feature vector using a multi-layered perceptron network for classifying based on feature vectors and provided with tags representing the user's maliciousness intentions.
15. A user classifying apparatus, comprising:
the target feature acquiring unit is used for acquiring a first individual feature of a target user, wherein the first individual feature is used for representing an individual attribute of the target user;
the meta-path traversing unit is used for traversing by using a meta-path method aiming at different relation types of the relation users of the target user to acquire a plurality of meta-paths each containing a plurality of relation users, and the meta-path method is used for acquiring the plurality of relation users similar based on different relation types in the relation users through similarity search;
a relation feature obtaining unit, configured to obtain a second feature and a relation feature of the relation user in each meta path, where the second feature is used to represent an individual attribute of the relation user, and the relation feature is used to represent a relation between the relation user and the target user;
an attention mechanism unit for using at least one attention mechanism based on the first individual feature, the second individual feature and the relationship feature of the stitching to obtain an attention feature vector, the at least one attention mechanism being used for calculating degrees of importance of different features to the target user, and the attention feature vector being a feature vector in which different features have different weights based on the degrees of importance; and
a classification result obtaining unit, configured to obtain a classification result of the target user based on the attention feature vector using a multi-layered sensor network, where the multi-layered sensor network is configured to perform classification based on the feature vector.
16. An overdue identification apparatus for financial users, comprising:
the financial user characteristic acquisition unit is used for acquiring a first individual characteristic of a financial user, and the first individual characteristic is used for representing an individual attribute of the financial user;
the financial meta-path traversing unit is used for traversing by using a meta-path method aiming at different relationship types of the relationship users of the financial users to acquire a plurality of meta-paths each containing a plurality of relationship users, and the meta-path method is used for acquiring the plurality of relationship users similar based on different relationship types in the relationship users through similarity search;
the financial relation characteristic acquisition unit is used for acquiring a second individual characteristic and a relation characteristic of the relation user in each meta path, wherein the second individual characteristic is used for representing individual attributes of the relation user, and the relation characteristic is used for representing the relation between the relation user and the financial user;
a financial attention mechanism unit for using at least one attention mechanism based on the first individual feature, the second individual feature and the relationship feature of the concatenation to obtain an attention feature vector, the at least one attention mechanism being used to calculate degrees of importance of different features to the target user, and the attention feature vector being a feature vector in which different features have different weights based on the degrees of importance; and
and the financial user overdue identification unit is used for identifying whether the financial user is overdue or not based on the attention characteristic vector by using a multi-layer sensor network, the multi-layer sensor network is used for classifying based on the characteristic vector and is provided with a label used for representing the overdue of the financial user.
17. A search recommendation apparatus for a target user, comprising:
the search user characteristic acquisition unit is used for acquiring a first individual characteristic of a target user, and the first individual characteristic is used for representing an individual attribute of the target user;
a search meta-path traversal unit, configured to traverse, using a meta-path method, for different relationship types of a relationship user of the target user, to obtain multiple meta-paths each including multiple relationship users, where the meta-path method is used to obtain, through similarity search, the multiple relationship users that are similar based on different relationship types among the relationship users;
a search relationship feature obtaining unit, configured to obtain a second feature and a relationship feature of the relationship user in each meta path, where the second feature is used to represent an individual attribute of the relationship user, and the relationship feature is used to represent a relationship between the relationship user and the target user;
a search attention mechanism unit for using at least one attention mechanism based on the first individual feature, the second individual feature and the relationship feature of the concatenation to obtain an attention feature vector, the at least one attention mechanism being used for calculating degrees of importance of different features to the target user, and the attention feature vector being a feature vector in which different features have different weights based on the degrees of importance;
the user classification unit is used for classifying the target user based on the attention feature vector by using a multilayer perceptron network, the multilayer perceptron network is used for classifying based on the feature vector and is provided with a label for expressing a category recommended to the target user for searching; and
and the search recommending unit is used for carrying out search recommendation on the target user according to the category.
18. A false transaction identification device, comprising:
the transaction user characteristic acquisition unit is used for acquiring a first individual characteristic of a transaction user, and the first individual characteristic is used for representing an individual attribute of the financial user;
the transaction meta-path traversing unit is used for traversing by using a meta-path method aiming at different relationship types of relationship users of the transaction users to acquire a plurality of meta-paths each including a plurality of relationship users, and the meta-path method is used for acquiring the plurality of relationship users similar based on different relationship types in the relationship users through similarity search;
the transaction relationship characteristic acquisition unit is used for acquiring a second characteristic and a relationship characteristic of the relationship user in each meta path, wherein the second characteristic is used for representing the individual attribute of the relationship user, and the relationship characteristic is used for representing the relationship between the relationship user and the transaction user;
a transaction attention mechanism unit for using at least one attention mechanism based on the first individual feature, the second individual feature and the relationship feature of the concatenation to obtain an attention feature vector, the at least one attention mechanism being used to calculate degrees of importance of different features to the target user, and the attention feature vector being a feature vector in which different features have different weights based on the degrees of importance; and
and the false transaction identification unit is used for identifying whether the transaction user carries out false transaction or not based on the attention characteristic vector by using a multi-layer sensor network, and the multi-layer sensor network is used for carrying out classification based on the characteristic vector and is provided with a label for representing false transaction.
19. An apparatus for identifying a user's intention to receive a coupon, comprising:
an intention user characteristic obtaining unit, configured to obtain a first individual characteristic of a user who picks up a coupon, the first individual characteristic being used for representing an individual attribute of the user;
the intention meta path traversing unit is used for traversing by using a meta path method aiming at different relation types of relation users of the users to acquire a plurality of meta paths each containing a plurality of relation users, and the meta path method is used for acquiring the plurality of relation users similar based on different relation types in the relation users through similarity search;
an intention relation characteristic obtaining unit, configured to obtain a second individual characteristic and a relation characteristic of the relation user in each meta path, where the second individual characteristic is used to represent an individual attribute of the relation user, and the relation characteristic is used to represent a relation between the relation user and the user receiving the coupon;
an intent attention mechanism unit for using at least one attention mechanism based on the first individual feature, the second individual feature and the relationship feature of the stitching to obtain an attention feature vector, the at least one attention mechanism being used to calculate degrees of importance of different features to the target user, and the attention feature vector being a feature vector in which different features have different weights based on the degrees of importance; and
a malicious intent recognition unit for recognizing whether the coupon picked up by the user is malicious intent based on the attention feature vector by using a multi-layer perceptron network, wherein the multi-layer perceptron network is used for classifying based on the feature vector and is provided with a label for representing the malicious intent of the user.
20. An electronic device, comprising:
a processor; and
memory having stored therein computer program instructions which, when executed by the processor, cause the processor to perform the method of user classification of any one of claims 1-10, the method of overdue identification of financial users of claim 11, the method of search recommendation of target users of claim 12, the method of false transaction identification of claim 13, and the method of identification of user's intention to receive a coupon of claim 14.
21. A computer readable medium having stored thereon computer program instructions which, when executed by a processor, cause the processor to perform the user categorization method of any one of claims 1 to 10, the financial user overdue identification method of claim 11, the search recommendation method of a target user of claim 12, the false transaction identification method of claim 13, and the identification method of the user's intention to receive a coupon of claim 14.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010001309.8A CN113065573B (en) | 2020-01-02 | 2020-01-02 | User classification method, user classification device and electronic equipment |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010001309.8A CN113065573B (en) | 2020-01-02 | 2020-01-02 | User classification method, user classification device and electronic equipment |
Publications (2)
Publication Number | Publication Date |
---|---|
CN113065573A true CN113065573A (en) | 2021-07-02 |
CN113065573B CN113065573B (en) | 2024-06-11 |
Family
ID=76558147
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202010001309.8A Active CN113065573B (en) | 2020-01-02 | 2020-01-02 | User classification method, user classification device and electronic equipment |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113065573B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113704566A (en) * | 2021-10-29 | 2021-11-26 | 贝壳技术有限公司 | Identification number body identification method, storage medium and electronic equipment |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20100125490A1 (en) * | 2008-11-14 | 2010-05-20 | Microsoft Corporation | Social network referral coupons |
CN108173746A (en) * | 2017-12-26 | 2018-06-15 | 东软集团股份有限公司 | Friend recommendation method, apparatus and computer equipment |
CN110046698A (en) * | 2019-04-28 | 2019-07-23 | 北京邮电大学 | Heterogeneous figure neural network generation method, device, electronic equipment and storage medium |
-
2020
- 2020-01-02 CN CN202010001309.8A patent/CN113065573B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20100125490A1 (en) * | 2008-11-14 | 2010-05-20 | Microsoft Corporation | Social network referral coupons |
CN108173746A (en) * | 2017-12-26 | 2018-06-15 | 东软集团股份有限公司 | Friend recommendation method, apparatus and computer equipment |
CN110046698A (en) * | 2019-04-28 | 2019-07-23 | 北京邮电大学 | Heterogeneous figure neural network generation method, device, electronic equipment and storage medium |
Non-Patent Citations (1)
Title |
---|
韩忠明;郑晨烨;段大高;董健;: "基于多信息融合表示学习的关联用户挖掘算法", 计算机科学, no. 04 * |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113704566A (en) * | 2021-10-29 | 2021-11-26 | 贝壳技术有限公司 | Identification number body identification method, storage medium and electronic equipment |
CN113704566B (en) * | 2021-10-29 | 2022-01-18 | 贝壳技术有限公司 | Identification number body identification method, storage medium and electronic equipment |
Also Published As
Publication number | Publication date |
---|---|
CN113065573B (en) | 2024-06-11 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
KR102433722B1 (en) | Information recommendation method, apparatus, device and medium | |
WO2019196546A1 (en) | Method and apparatus for determining risk probability of service request event | |
CN112084383A (en) | Information recommendation method, device and equipment based on knowledge graph and storage medium | |
US11682093B2 (en) | Document term recognition and analytics | |
WO2018040944A1 (en) | System, method, and device for identifying malicious address/malicious purchase order | |
JP6345203B2 (en) | Program, system, and method for determining similarity of objects | |
CN111737546B (en) | Method and device for determining entity service attribute | |
US20210192496A1 (en) | Digital wallet reward optimization using reverse-engineering | |
US20230153581A1 (en) | Artificial intelligence system employing graph convolutional networks for analyzing multi-entity-type multi-relational data | |
JP2022548501A (en) | Data acquisition method and device for analyzing cryptocurrency transactions | |
CN111429161A (en) | Feature extraction method, feature extraction device, storage medium, and electronic apparatus | |
CN114637850A (en) | Abnormal behavior recognition and model training method, device, equipment and storage medium | |
CN113065573B (en) | User classification method, user classification device and electronic equipment | |
CN114330476A (en) | Model training method for media content recognition and media content recognition method | |
KR102563095B1 (en) | AI-based open market integrated management system | |
CN113779276A (en) | Method and device for detecting comments | |
JP2021197089A (en) | Output device, output method, and output program | |
Elzeheiry et al. | Sentiment analysis for e-commerce product reviews: Current trends and future directions | |
JP2003323601A (en) | Predicting device with reliability scale | |
CN112115258B (en) | Credit evaluation method and device for user, server and storage medium | |
CN115187252A (en) | Method for identifying fraud in network transaction system, server and storage medium | |
JP2018067215A (en) | Data analysis system, control method thereof, program, and recording medium | |
JP2018160256A (en) | Program, system, and method for determining similarity of object | |
JP7302107B1 (en) | LEARNING SYSTEMS, LEARNING METHODS AND PROGRAMS | |
KR102554886B1 (en) | Open market data analysis-based advertising efficiency maximization platform |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |