CN105335519B - Model generation method and device and recommendation method and device - Google Patents
Model generation method and device and recommendation method and device Download PDFInfo
- Publication number
- CN105335519B CN105335519B CN201510794561.8A CN201510794561A CN105335519B CN 105335519 B CN105335519 B CN 105335519B CN 201510794561 A CN201510794561 A CN 201510794561A CN 105335519 B CN105335519 B CN 105335519B
- Authority
- CN
- China
- Prior art keywords
- entity
- user
- feature vector
- obtaining
- vector
- 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.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9535—Search customisation based on user profiles and personalisation
Landscapes
- Engineering & Computer Science (AREA)
- Databases & Information Systems (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The embodiment of the invention provides a model generation method and device and a recommendation method and device. On one hand, in the model generation method of the embodiment of the invention, at least one of the document content characteristic vector of each entity, the logic association relation characteristic vector between each entity, the user behavior relation characteristic vector of each entity and the characteristic vector of each entity in the knowledge graph is obtained; and performing machine learning according to at least one of the document content feature vector, the logic association relation feature vector, the user behavior relation feature vector and the feature vector to generate a deep fusion model. Therefore, the technical scheme provided by the embodiment of the invention can generate the deep fusion model by integrating various relationships among the entities, and the deep fusion model can be used for obtaining the surprise among the entities, so that the entities can be recommended to the user based on the surprise, the search recommendation requirements of the user are met, and the click rate of the recommended entities is improved.
Description
[ technical field ] A method for producing a semiconductor device
The invention relates to the technical field of search, in particular to a model generation method and device and a recommendation method and device.
[ background of the invention ]
Currently, search recommendations are based on motivating the potential needs of the user by providing the user with other content that may be of interest in relation to the query terms, if the user's primary search needs are met. For example, referring to fig. 1, which is a first exemplary diagram of prior art knowledge-graph based search recommendations, as shown, when a user queries "university of princeton", a famous alumni of princeton university shown in fig. 1 can be recommended in a non-search result area of a search result page, which is a recommending entity that is very related to the query word "university of princeton".
However, in the prior art, when search recommendation is performed based on the knowledge graph, the recommended entities are generally known and cannot arouse the user interest. Therefore, the search recommendation method cannot meet the search recommendation requirements of the user, and the click rate of the recommending entity is low.
[ summary of the invention ]
In view of this, embodiments of the present invention provide a model generation method and apparatus, and a recommendation method and apparatus, in which a deep fusion model is generated by integrating various relationships between entities, and the deep fusion model can be used to obtain surprise between entities, so that entities can be recommended to a user based on the surprise, the search recommendation requirements of the user are met, and the click rate of recommended entities is improved.
In one aspect of the embodiments of the present invention, a model generation method is provided, including:
obtaining at least one of document content feature vectors of all entities in a knowledge graph, logic association relation feature vectors among all the entities, user behavior relation feature vectors of all the entities and feature vectors of all the entities;
and performing machine learning according to at least one of the document content feature vector, the logic association relation feature vector, the user behavior relation feature vector and the feature vector to generate a depth fusion model.
The above-described aspect and any possible implementation manner further provide an implementation manner, where obtaining the user behavior relationship feature vector of each entity includes:
acquiring a historical search behavior record of a user;
obtaining the searching behavior and clicking behavior of the user aiming at each entity according to the historical searching behavior record;
and obtaining the user behavior relation characteristic vector of each entity according to the search behavior and click behavior of the user aiming at each entity.
The above-described aspect and any possible implementation manner further provide an implementation manner, where obtaining the feature vector of each entity includes: and randomly generating the characteristic vector for each entity according to the entities defined in the knowledge graph.
In one aspect of the embodiments of the present invention, a recommendation method is provided, including:
obtaining a candidate entity corresponding to the designated entity;
inputting at least one of a document content feature vector of the specified entity, a logic association relation feature vector between the specified entity and a candidate entity, a user behavior relation feature vector of the specified entity and a feature vector of the specified entity, and at least one of a document content feature vector of the candidate entity and a feature vector of the candidate entity into a depth fusion model to obtain surprise of the candidate entity; the depth fusion model is obtained by using the method of one of claims 1 to 3;
and obtaining a recommended entity corresponding to the specified entity according to the surprise and the candidate entity.
The above-described aspect and any possible implementation manner further provide an implementation manner, where the obtaining of the candidate entity corresponding to the specified entity includes:
obtaining the candidate entity according to the designated entity and the user behavior relation characteristic vector of the designated entity; or obtaining the candidate entity according to the entity defined in the knowledge graph.
The above-mentioned aspect and any possible implementation manner further provide an implementation manner, where obtaining, according to the surprise and the candidate entity, a recommended entity corresponding to the specified entity includes:
and sequencing the candidate entities according to the sequence of the surprise degrees from big to small to obtain a sequencing result, and taking at least one candidate entity which is sequenced at the front in the sequencing result as a recommending entity corresponding to the specified entity.
The above-described aspects and any possible implementations further provide an implementation, and the method further includes:
and adjusting the feature vector of the specified entity according to the recommended entity corresponding to the specified entity, wherein the feature vector obtained after adjustment is used for generating the depth fusion model.
In one aspect of the embodiments of the present invention, a model generation apparatus is provided, including:
the system comprises a vector acquisition unit, a knowledge graph analysis unit and a knowledge graph analysis unit, wherein the vector acquisition unit is used for acquiring at least one of document content characteristic vectors of all entities in the knowledge graph, logic association relation characteristic vectors among the entities, user behavior relation characteristic vectors of the entities and characteristic vectors of the entities;
and the model generating unit is used for performing machine learning according to at least one of the document content characteristic vector, the logic association relation characteristic vector, the user behavior relation characteristic vector and the characteristic vector to generate a deep fusion model.
As for the above-mentioned aspect and any possible implementation manner, there is further provided an implementation manner, where the vector obtaining unit is specifically configured to:
acquiring a historical search behavior record of a user;
obtaining the searching behavior and clicking behavior of the user aiming at each entity according to the historical searching behavior record;
and obtaining the user behavior relation characteristic vector of each entity according to the search behavior and click behavior of the user aiming at each entity.
As for the above-mentioned aspect and any possible implementation manner, there is further provided an implementation manner, where the vector obtaining unit is specifically configured to: and randomly generating the characteristic vector for each entity according to the entities defined in the knowledge graph.
In one aspect of the embodiments of the present invention, a recommendation apparatus is provided, including:
an entity obtaining unit, configured to obtain a candidate entity corresponding to a specified entity;
a surprise obtaining unit, configured to input at least one of the document content feature vector of the specified entity, the logic association relationship feature vector between the specified entity and the candidate entity, the user behavior relationship feature vector of the specified entity, and the feature vector of the specified entity, and at least one of the document content feature vector of the candidate entity and the feature vector of the candidate entity into a deep fusion model to obtain a surprise of the candidate entity; the depth fusion model is generated using the apparatus of one of claims 8 to 10;
and the entity processing unit is used for obtaining a recommended entity corresponding to the specified entity according to the surprise and the candidate entity.
As for the above-mentioned aspect and any possible implementation manner, there is further provided an implementation manner, where the entity obtaining unit is specifically configured to:
obtaining the candidate entity according to the designated entity and the user behavior relation characteristic vector of the designated entity; or obtaining the candidate entity according to the entity defined in the knowledge graph.
The above-mentioned aspect and any possible implementation further provide an implementation, where the entity processing unit is specifically configured to:
and sequencing the candidate entities according to the sequence of the surprise degrees from big to small to obtain a sequencing result, and taking at least one candidate entity which is sequenced at the front in the sequencing result as a recommending entity corresponding to the specified entity.
The above-described aspects and any possible implementations further provide an implementation, where the apparatus further includes:
and the vector adjusting unit is used for adjusting the feature vector of the specified entity according to the recommended entity corresponding to the specified entity, and the feature vector obtained after adjustment is used for generating the depth fusion model.
According to the technical scheme, the embodiment of the invention has the following beneficial effects:
the technical scheme provided by the embodiment of the invention can generate the deep fusion model by integrating various relationships among the entities, and the deep fusion model can be used for obtaining the surprise among the entities, so that the entities can be recommended to the user based on the surprise. Compared with the mode of searching and recommending only based on the knowledge graph in the prior art, the recommending entity provided by the embodiment of the invention can arouse the interest of the user, so that the searching and recommending requirements of the user can be met, and the click rate of the recommending entity is improved.
[ description of the drawings ]
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without inventive labor.
FIG. 1 is a first exemplary prior art knowledge graph-based search recommendation;
FIG. 2 is a schematic flow chart diagram of a model generation method provided by an embodiment of the invention;
FIG. 3 is a diagram of an exemplary generation of a depth fusion model provided by an embodiment of the invention;
FIG. 4 is a flowchart illustrating a recommendation method according to an embodiment of the present invention;
FIG. 5 is a first exemplary diagram of a deep fusion model based search recommendation provided by an embodiment of the present invention;
FIG. 6 is a second exemplary prior art knowledge-graph based search recommendation;
FIG. 7 is a diagram illustrating a second example of search recommendation based on a depth fusion model according to an embodiment of the present invention;
FIG. 8 is a functional block diagram of a model generation apparatus according to an embodiment of the present invention;
fig. 9 is a functional block diagram of a recommendation device according to an embodiment of the present invention.
[ detailed description ] embodiments
For better understanding of the technical solutions of the present invention, the following detailed descriptions of the embodiments of the present invention are provided with reference to the accompanying drawings.
It should be understood that the described embodiments are only some embodiments of the invention, and not all embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The terminology used in the embodiments of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. As used in the examples of the present invention and the appended claims, the singular forms "a," "an," and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise.
It should be understood that the term "and/or" as used herein is merely one type of association that describes an associated object, meaning that three relationships may exist, e.g., a and/or B may mean: a exists alone, A and B exist simultaneously, and B exists alone. In addition, the character "/" herein generally indicates that the former and latter related objects are in an "or" relationship.
The word "if" as used herein may be interpreted as "at … …" or "when … …" or "in response to a determination" or "in response to a detection", depending on the context. Similarly, the phrases "if determined" or "if detected (a stated condition or event)" may be interpreted as "when determined" or "in response to a determination" or "when detected (a stated condition or event)" or "in response to a detection (a stated condition or event)", depending on the context.
Example one
An embodiment of the present invention provides a model generation method, please refer to fig. 2, which is a schematic flow chart of the model generation method provided in the embodiment of the present invention, and as shown in the figure, the method includes the following steps:
s201, obtaining at least one of document content characteristic vectors of all entities in the knowledge graph, logic association relation characteristic vectors among all the entities, user behavior relation characteristic vectors of all the entities and characteristic vectors of all the entities.
S202, performing machine learning according to at least one of the document content feature vector, the logic association relation feature vector, the user behavior relation feature vector and the feature vector to generate a depth fusion model.
It should be noted that each entity and the related information of each entity are defined in the knowledge graph; the entity refers to a real life thing such as a person, an article, a virtual person or a place, etc.
Referring to fig. 3, which is a diagram illustrating a generation example of a deep fusion model according to an embodiment of the present invention, as shown in the diagram, before generating the deep fusion model, for any two entities e1 and e2 in a knowledge graph, a document content feature vector s1 of an entity e1, a document content feature vector s2 of the entity e2, a logical association relationship feature vector k between an entity e1 and an entity e2, a user behavior relationship feature vector c between an entity e1 and an entity e2, a feature vector p1 of the entity e1, and a feature vector p2 of the entity e2 need to be obtained.
For example, in the embodiment of the present invention, the method for obtaining the document content feature vector of each entity may include, but is not limited to: document d1 of entity e1 and document d2 of entity e2 may be modeled using convolutional neural networks to obtain document content feature vector s1 for entity e1 and document content feature vector s2 for entity e 2.
For example, taking entity e1 as an example: document d1 of entity e1 may be obtained from the knowledge graph first, e.g., the text in the encyclopedia of entity e1 may be taken as document d1 of entity e 1. Then, word feature vectors w 1-wn are extracted from the document d 1. Then, the word feature vectors w 1-wn are convolved in the convolutional layer to obtain vector features. Finally, the vector feature output by the convolutional layer is subjected to maximum pooling processing at the maximum pooling layer to obtain a document content feature vector s1 of the entity e 1. The method comprises the steps that a convolution model used in convolution operation and document content feature vectors of an entity can be automatically optimized in the process of deep machine training based on a deep neural network when a deep fusion model is generated.
For example, in the embodiment of the present invention, the method for obtaining the feature vector of the logical association relationship between the entities may include, but is not limited to: the logical association relationship feature vector k between entity e1 and entity e2 may be obtained from a knowledge graph. It is to be appreciated that the logical association feature vector k can represent a logical association between entity e1 and entity e2 in the knowledge-graph.
For example, in the embodiment of the present invention, the method for obtaining the user behavior relationship vector may include, but is not limited to:
first, a history search behavior record of a user is obtained. And then, obtaining the searching behavior and clicking behavior of the user aiming at each entity according to the historical searching behavior record of the user. And finally, obtaining the user behavior relation characteristic vector of each entity according to the search behavior and click behavior of the user aiming at each entity.
It can be understood that, in the embodiment of the present invention, the user behavior relationship feature vector of each entity includes a plurality of numerical values, each numerical value may represent a user behavior relationship between the entity and another entity, so the user behavior relationship feature vector may also be understood as a user behavior relationship feature vector between one entity and another entity.
For example, after the user has searched the entity e1 in the search engine, and clicked the entity e2 in the recommended entity on the right side of the search result page, the number of clicks in the user behavior relationship feature vector of the entity e1 and the entity e2 is increased by 1. After the user searches the entity e1 in the search engine, searching the entity e2 in the search engine, and adding 1 to the search times in the user behavior relation feature vector of the entity e1 and the entity e 2; and after the user searches the entity e1 in the search engine, if the clicked search result contains the information of another entity e2, the number of hops in the user behavior relation feature vector of the entity e1 and the entity e2 is increased by 1. It can be understood that the above statistical manner for obtaining the numerical values in the user behavior relationship feature vector is only an example, and in the embodiment of the present invention, the manner for obtaining the user behavior relationship feature vector of each entity according to the search behavior and the click behavior of the user for each entity is not particularly limited.
In the embodiment of the invention, the feature vector of the entity comprises other entities which have lower relevance with the entity but can cause surprise of the user, and has no particularly obvious association relation with the entity.
For example, in the embodiment of the present invention, the method for obtaining the feature vector of each entity may include, but is not limited to:
the feature vectors may be randomly generated for each entity based on the entities defined in the knowledge-graph. Or after a depth fusion model is generated according to randomly generated feature vectors, a corresponding recommended entity is obtained for a specified entity by using the depth fusion model, then the feature vectors of the specified entity are adjusted according to the recommended entity corresponding to the specified entity, and then machine learning is carried out again by using the adjusted feature vectors to generate a new depth fusion model, so that continuous optimization of the feature vectors and the depth fusion model is realized. Or, the feature vector of the specified entity can be adjusted through a back propagation optimization mechanism of training errors in the process of deep machine learning by using a deep neural network.
In a specific implementation process, as shown in fig. 3, at least one of the obtained document content feature vector, the logic association relationship feature vector, the user behavior relationship feature vector, and the feature vector may be input to a deep neural network, and the deep neural network performs deep machine learning on user preferences according to the input vector to generate a deep fusion model.
Example two
The embodiment of the invention provides a recommendation method, and the depth fusion model used in the recommendation method provided by the embodiment is the depth fusion model generated by using the model generation method provided by the first embodiment. Please refer to fig. 4, which is a flowchart illustrating a recommendation method according to an embodiment of the present invention, wherein the method includes the following steps:
s401, obtaining a candidate entity corresponding to the specified entity.
S402, inputting at least one of the document content characteristic vector of the specified entity, the logic association relation characteristic vector between the specified entity and the candidate entity, the user behavior relation characteristic vector of the specified entity and the characteristic vector of the specified entity, and at least one of the document content characteristic vector of the candidate entity and the characteristic vector of the candidate entity into a depth fusion model to obtain the surprise of the candidate entity; the depth fusion model is obtained by using the model generation method.
And S403, obtaining a recommended entity corresponding to the specified entity according to the surprise and the candidate entity.
For example, in the embodiment of the present invention, the method for obtaining the candidate entity corresponding to the specified entity may include, but is not limited to, the following two methods:
the first method comprises the following steps: and obtaining the candidate entity according to the entity defined in the knowledge graph. For example, all entities defined in the knowledge-graph may be considered as the candidate entities.
And the second method comprises the following steps: according to the name of the designated entity input by the user, obtaining a plurality of user behavior relation characteristic vectors of the designated entity; and then, obtaining the candidate entity according to the specified entity and the user behavior relation characteristic vector of the specified entity.
It can be understood that, in the first embodiment, the user behavior relationship feature vectors of the entities may be stored after being obtained, so that, when a recommended entity needs to be obtained by using the deep fusion model, a plurality of user behavior relationship feature vectors corresponding to the specified entity may be found according to the name of the specified entity, where the user behavior relationship feature vectors represent user behavior relationships between the specified entity and each entity in the plurality of entities. In the method, a plurality of entities with user behavior relation feature vectors between the entities and the designated entity can be used as candidate entities corresponding to the designated entity.
Compared with the first method, the second method utilizes the user behavior relation feature vector to screen the entities defined in the knowledge graph, so as to narrow the range of the candidate entities and reduce the calculation amount when the deep fusion model is utilized to obtain the recommended entities, thereby improving the efficiency of the deep fusion model in obtaining the recommended entities.
In the embodiment of the present invention, when obtaining the surprise of each candidate entity, at least one of the document content feature vector of the designated entity, the logic association relationship feature vector between the designated entity and the candidate entity, the user behavior relationship feature vector of the designated entity, and the feature vector of the designated entity, and at least one of the document content feature vector of the candidate entity and the feature vector of the candidate entity may be input into the deep fusion model, and the deep fusion model may calculate and output the surprise of each candidate entity, thereby obtaining the surprise of the candidate entity.
Here, the vector to be input to the depth fusion model needs to be identical to the vector used in the deep machine learning when the depth fusion model is generated, and for example, when the depth machine learning is performed by using the logical association relationship feature vector between the designated entity and the candidate entity when the depth fusion model is generated, the logical association relationship feature vector between the designated entity and the candidate entity needs to be input to the depth fusion model. Alternatively, for example, when performing deep machine learning using the feature vector of the entity in generating the deep fusion model, it is necessary to input the feature vector of the specified entity and the feature vector of the candidate entity into the deep fusion model.
It is understood that the expectation degree refers to the proportion of recommending entities in the recommending result, which are generated by the knowledge graph and the rule, in all recommending entities. The surprise degree is equal to 1 minus the expectation degree, and refers to the proportion of other entities except the recommending entities generated by the knowledge graph and the rule in the recommending result to all recommending entities, and is the predicted surprise degree of the user for the recommending entities when the recommending entities are provided for the user after the name of the specified entity is input by the user.
For example, the method for obtaining the recommended entity corresponding to the first entity according to the surprise and the candidate entity may include, but is not limited to:
firstly, the candidate entities are ranked according to the sequence of the surprise degrees from big to small so as to obtain a ranking result. And then, extracting at least one candidate entity with the corresponding number in the top ranking order from the ranking result according to the preset recommendation number, and taking the extracted at least one candidate entity as the recommendation entity corresponding to the specified entity.
It will be appreciated that the obtained recommending entities may be recommended to the user when outputting search results to the user that match the specified entity, e.g., the recommending entities may be presented to the right of the search results page.
Optionally, in a possible implementation manner of this embodiment, the feature vector of the specified entity may be further adjusted according to the recommended entity corresponding to the specified entity, and the entity included in the feature vector of the specified entity may be one or more of the recommended entities. Further, the feature vector obtained after adjustment can be used for performing deep machine learning to generate a new deep fusion model, the new deep fusion model can be further used for obtaining a recommended entity, and so on, so that the feature vector of the entity can be continuously optimized and adjusted, and the deep fusion model can be optimized and adjusted repeatedly, thereby continuously improving the obtaining accuracy of the recommended entity, continuously improving the satisfaction degree of a user on the recommended entity, and improving the click rate of the recommended entity.
For example, referring to fig. 5, which is a first exemplary diagram of a search recommendation based on a deep fusion model according to an embodiment of the present invention, as shown in fig. 5, if a name of a specific entity input by a user is "university of princeton", if a search recommendation based on a knowledge graph in the prior art is used, the recommendation entities shown in fig. 1 are obtained, and are well known to the user and cannot arouse the interest of the user. However, by using the deep fusion model provided by the embodiment of the invention, the recommending entities shown in fig. 5 can be obtained, and the recommending entities are entities which are not obviously related to the specified entity but some scholars, and obviously the scholars can arouse the interest of the user, so that the click of the user is triggered, the potential search requirement of the user is stimulated, and therefore the recommending entities can better meet the requirement of the user, and the recommending accuracy and the click rate of the recommending entities are improved.
For another example, please refer to fig. 6 and fig. 7, which are a second exemplary diagram of a search recommendation based on a knowledge graph in the prior art and a second exemplary diagram of a search recommendation based on a depth fusion model provided in an embodiment of the present invention, respectively.
When a user searches for "Halloween," the user's potential needs may include: a thriller movie, a halloween-related movie, props needed to prepare the halloween, a game/theme of the halloween party, other horrors/monsters/creatures. As shown in fig. 6, the recommending entity to be displayed to the user based on the knowledge graph is the midwest festival related to Halloween, which brings a low surprise to the user. However, as shown in fig. 7, if the recommending entity generated by the deep fusion model contains 7 horror movies (the entity identified by the dashed box in fig. 7) and all 5 other recommending entities (the entity identified by the implementation box in fig. 7), the coverage of these recommending entities is wider, and the recommending entity shown in fig. 7 brings more surprise to the user. The recommended entities indicated by solid and dashed boxes in FIG. 7 represent entities with high user click rates in the experiment. It can be seen that the surprise of the entity mined by the deep fusion model does gain more attention and interest of the user.
The embodiment of the invention further provides an embodiment of a device for realizing the steps and the method in the embodiment of the method.
Please refer to fig. 8, which is a functional block diagram of a model generation apparatus according to an embodiment of the present invention. As shown, the apparatus comprises:
the vector acquiring unit 81 is configured to acquire at least one of a document content feature vector of each entity, a logic association relationship feature vector between the entities, a user behavior relationship feature vector of each entity, and a feature vector of each entity in the knowledge graph;
and the model generating unit 82 is configured to perform machine learning according to at least one of the document content feature vector, the logic association relationship feature vector, the user behavior relationship feature vector, and the feature vector, and generate a deep fusion model.
In a specific implementation process, the vector obtaining unit 81 is specifically configured to:
acquiring a historical search behavior record of a user;
obtaining the searching behavior and clicking behavior of the user aiming at each entity according to the historical searching behavior record;
and obtaining the user behavior relation characteristic vector of each entity according to the search behavior and click behavior of the user aiming at each entity.
In a specific implementation process, the vector obtaining unit 81 is specifically configured to: and randomly generating the characteristic vector for each entity according to the entities defined in the knowledge graph.
Since each unit in the present embodiment can execute the method shown in fig. 2, reference may be made to the related description of fig. 2 for a part of the present embodiment that is not described in detail.
Please refer to fig. 9, which is a block diagram of a recommendation device according to an embodiment of the present invention. As shown, the apparatus comprises:
an entity obtaining unit 91, configured to obtain a candidate entity corresponding to the specified entity;
a surprise obtaining unit 92, configured to input at least one of the document content feature vector of the specified entity, the logic association relationship feature vector between the specified entity and the candidate entity, the user behavior relationship feature vector of the specified entity, and the feature vector of the specified entity, and at least one of the document content feature vector of the candidate entity and the feature vector of the candidate entity into a deep fusion model to obtain a surprise of the candidate entity; the depth fusion model is generated by using a model generation device;
and the entity processing unit 93 is configured to obtain a recommended entity corresponding to the specified entity according to the surprise and the candidate entity.
In a specific implementation process, the entity obtaining unit 91 is specifically configured to:
obtaining the candidate entity according to the designated entity and the user behavior relation characteristic vector of the designated entity; or obtaining the candidate entity according to the entity defined in the knowledge graph.
In a specific implementation process, the entity processing unit 93 is specifically configured to:
and sequencing the candidate entities according to the sequence of the surprise degrees from big to small to obtain a sequencing result, and taking at least one candidate entity which is sequenced at the front in the sequencing result as a recommending entity corresponding to the specified entity.
Optionally, in a possible implementation manner of this embodiment, the apparatus further includes:
a vector adjusting unit 94, configured to adjust a feature vector of the specified entity according to the recommended entity corresponding to the specified entity, where the feature vector obtained after the adjustment is used to generate the depth fusion model.
Since each unit in the present embodiment can execute the method shown in fig. 4, reference may be made to the related description of fig. 4 for a part of the present embodiment that is not described in detail.
The technical scheme of the embodiment of the invention has the following beneficial effects:
in the embodiment of the invention, document content vectors of all entities in a knowledge graph, logic association relation vectors among all the entities, user behavior relation vectors of all the entities and characteristic vectors of all the entities are obtained; and performing machine learning according to the document content vector, the logic association relation vector, the user behavior relation vector and the feature vector to generate a deep fusion model.
The technical scheme provided by the embodiment of the invention can generate the deep fusion model by integrating various relationships among the entities, and the deep fusion model can be used for obtaining the surprise among the entities, so that the entities can be recommended to the user based on the surprise. Compared with the mode of searching and recommending only based on the knowledge graph in the prior art, the recommending entity provided by the embodiment of the invention can arouse the interest of the user, so that the searching and recommending requirements of the user can be met, and the click rate of the recommending entity is improved.
It is clear to those skilled in the art that, for convenience and brevity of description, the specific working processes of the above-described systems, apparatuses and units may refer to the corresponding processes in the foregoing method embodiments, and are not described herein again.
In the embodiments provided in the present invention, it should be understood that the disclosed system, apparatus and method may be implemented in other ways. For example, the above-described apparatus embodiments are merely illustrative, and for example, the division of the units is only one logical division, and there may be other divisions in actual implementation, for example, a plurality of units or components may be combined or integrated into another system, or some features may be omitted, or not executed. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be an indirect coupling or communication connection through some interfaces, devices or units, and may be in an electrical, mechanical or other form.
The units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of the embodiment.
In addition, functional units in the embodiments of the present invention may be integrated into one processing unit, or each unit may exist alone physically, or two or more units are integrated into one unit. The integrated unit can be realized in a form of hardware, or in a form of hardware plus a software functional unit.
The integrated unit implemented in the form of a software functional unit may be stored in a computer readable storage medium. The software functional unit is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device) or a Processor (Processor) to execute some steps of the methods according to the embodiments of the present invention. And the aforementioned storage medium includes: various media capable of storing program codes, such as a usb disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk, or an optical disk.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents, improvements and the like made within the spirit and principle of the present invention should be included in the scope of the present invention.
Claims (14)
1. A method of model generation, the method comprising:
obtaining at least one of document content feature vectors of all entities in a knowledge graph, logic association relation feature vectors among all the entities, user behavior relation feature vectors of all the entities and feature vectors of all the entities; the user behavior relation characteristic vector of each entity is obtained according to the searching behavior and the clicking behavior of the user aiming at each entity in the historical searching behavior record of the user;
performing machine learning according to at least one of the document content feature vector, the logic association relation feature vector, the user behavior relation feature vector and the feature vector to generate a depth fusion model; the depth fusion model is used for determining a recommended entity for a specified entity input by a user when search recommendation is performed.
2. The method of claim 1, wherein obtaining the user behavior relationship feature vector for each entity comprises:
acquiring a historical search behavior record of a user;
obtaining the searching behavior and clicking behavior of the user aiming at each entity according to the historical searching behavior record;
and obtaining the user behavior relation characteristic vector of each entity according to the search behavior and click behavior of the user aiming at each entity.
3. The method of claim 1, wherein obtaining the feature vector for each entity comprises: and randomly generating the characteristic vector for each entity according to the entities defined in the knowledge graph.
4. A recommendation method, characterized in that the method comprises:
obtaining a candidate entity corresponding to a specified entity input by a user during searching;
inputting at least one of a document content feature vector of the specified entity, a logic association relation feature vector between the specified entity and a candidate entity, a user behavior relation feature vector of the specified entity and a feature vector of the specified entity, and at least one of a document content feature vector of the candidate entity and a feature vector of the candidate entity into a depth fusion model to obtain surprise of the candidate entity; the depth fusion model is obtained by using the method of one of claims 1 to 3;
and obtaining a recommended entity corresponding to the specified entity according to the surprise and the candidate entity.
5. The method of claim 4, wherein the obtaining of the candidate entity corresponding to the specified entity input by the user in the search comprises:
obtaining the candidate entity according to the designated entity and the user behavior relation characteristic vector of the designated entity; or obtaining the candidate entity according to the entity defined in the knowledge graph.
6. The method of claim 4, wherein obtaining the recommended entity corresponding to the designated entity according to the surprise and the candidate entity comprises:
and sequencing the candidate entities according to the sequence of the surprise degrees from big to small to obtain a sequencing result, and taking at least one candidate entity which is sequenced at the front in the sequencing result as a recommending entity corresponding to the specified entity.
7. The method according to any one of claims 4 to 6, further comprising:
and adjusting the feature vector of the specified entity according to the recommended entity corresponding to the specified entity, wherein the feature vector obtained after adjustment is used for generating the depth fusion model.
8. An apparatus for model generation, the apparatus comprising:
the system comprises a vector acquisition unit, a knowledge graph analysis unit and a knowledge graph analysis unit, wherein the vector acquisition unit is used for acquiring at least one of document content characteristic vectors of all entities in the knowledge graph, logic association relation characteristic vectors among the entities, user behavior relation characteristic vectors of the entities and characteristic vectors of the entities; the user behavior relation characteristic vector of each entity is obtained according to the searching behavior and the clicking behavior of the user aiming at each entity in the historical searching behavior record of the user;
the model generating unit is used for performing machine learning according to at least one of the document content characteristic vector, the logic association relation characteristic vector, the user behavior relation characteristic vector and the characteristic vector to generate a deep fusion model; the depth fusion model is used for determining a recommended entity for a specified entity input by a user when search recommendation is performed.
9. The apparatus according to claim 8, wherein the vector obtaining unit is specifically configured to:
acquiring a historical search behavior record of a user;
obtaining the searching behavior and clicking behavior of the user aiming at each entity according to the historical searching behavior record;
and obtaining the user behavior relation characteristic vector of each entity according to the search behavior and click behavior of the user aiming at each entity.
10. The apparatus according to claim 8, wherein the vector obtaining unit is specifically configured to: and randomly generating the characteristic vector for each entity according to the entities defined in the knowledge graph.
11. A recommendation device, characterized in that the device comprises:
the entity acquisition unit is used for acquiring a candidate entity corresponding to a specified entity input by a user during searching;
a surprise obtaining unit, configured to input at least one of the document content feature vector of the specified entity, the logic association relationship feature vector between the specified entity and the candidate entity, the user behavior relationship feature vector of the specified entity, and the feature vector of the specified entity, and at least one of the document content feature vector of the candidate entity and the feature vector of the candidate entity into a deep fusion model to obtain a surprise of the candidate entity; the depth fusion model is generated using the apparatus of one of claims 8 to 10;
and the entity processing unit is used for obtaining a recommended entity corresponding to the specified entity according to the surprise and the candidate entity.
12. The apparatus of claim 11, wherein the entity obtaining unit is specifically configured to:
obtaining the candidate entity according to the designated entity and the user behavior relation characteristic vector of the designated entity; or obtaining the candidate entity according to the entity defined in the knowledge graph.
13. The apparatus according to claim 11, wherein the entity processing unit is specifically configured to:
and sequencing the candidate entities according to the sequence of the surprise degrees from big to small to obtain a sequencing result, and taking at least one candidate entity which is sequenced at the front in the sequencing result as a recommending entity corresponding to the specified entity.
14. The apparatus of any one of claims 11 to 13, further comprising:
and the vector adjusting unit is used for adjusting the feature vector of the specified entity according to the recommended entity corresponding to the specified entity, and the feature vector obtained after adjustment is used for generating the depth fusion model.
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510794561.8A CN105335519B (en) | 2015-11-18 | 2015-11-18 | Model generation method and device and recommendation method and device |
| PCT/CN2016/089648 WO2017084362A1 (en) | 2015-11-18 | 2016-07-11 | Model generation method, recommendation method and corresponding apparatuses, device and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510794561.8A CN105335519B (en) | 2015-11-18 | 2015-11-18 | Model generation method and device and recommendation method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105335519A CN105335519A (en) | 2016-02-17 |
| CN105335519B true CN105335519B (en) | 2021-08-17 |
Family
ID=55286046
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510794561.8A Active CN105335519B (en) | 2015-11-18 | 2015-11-18 | Model generation method and device and recommendation method and device |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN105335519B (en) |
| WO (1) | WO2017084362A1 (en) |
Families Citing this family (33)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105335519B (en) * | 2015-11-18 | 2021-08-17 | 百度在线网络技术（北京）有限公司 | Model generation method and device and recommendation method and device |
| CN105808931B (en) * | 2016-03-03 | 2019-05-07 | 北京大学深圳研究生院 | A kind of the acupuncture decision support method and device of knowledge based map |
| CN105912500B (en) | 2016-03-30 | 2017-11-14 | 百度在线网络技术（北京）有限公司 | Machine learning model generation method and device |
| CN106022877A (en) * | 2016-05-19 | 2016-10-12 | 华南理工大学 | User mobile game behavior map-based game recommendation method |
| CN106649774A (en) * | 2016-12-27 | 2017-05-10 | 北京百度网讯科技有限公司 | Artificial intelligence-based object pushing method and apparatus |
| CN108804387B (en) * | 2017-04-27 | 2021-07-23 | 腾讯科技（深圳）有限公司 | Target user determination method and device |
| CN107330050A (en) * | 2017-06-28 | 2017-11-07 | 达而观信息科技（上海）有限公司 | A kind of personalized recommendation method and system |
| CN109388793B (en) * | 2017-08-03 | 2023-04-07 | 阿里巴巴集团控股有限公司 | Entity marking method, intention identification method, corresponding device and computer storage medium |
| CN107463701B (en) * | 2017-08-15 | 2020-09-08 | 北京百度网讯科技有限公司 | Method and device for pushing information stream based on artificial intelligence |
| CN107506486A (en) * | 2017-09-21 | 2017-12-22 | 北京航空航天大学 | A kind of relation extending method based on entity link |
| TWI651664B (en) * | 2017-11-15 | 2019-02-21 | 財團法人資訊工業策進會 | Model generation server and its model generation method |
| CN108509479B (en) * | 2017-12-13 | 2022-02-11 | 深圳市腾讯计算机***有限公司 | Entity recommendation method and device, terminal and readable storage medium |
| CN108334588B (en) * | 2018-01-29 | 2021-03-23 | 北京搜狐新媒体信息技术有限公司 | User tag construction method and device |
| CN108345702A (en) * | 2018-04-10 | 2018-07-31 | 北京百度网讯科技有限公司 | Entity recommends method and apparatus |
| CN108665394B (en) * | 2018-05-07 | 2021-11-30 | 北京文华在线教育科技股份有限公司 | Self-adaptive learning method and system |
| CN108829854B (en) * | 2018-06-21 | 2021-08-31 | 北京百度网讯科技有限公司 | Method, apparatus, device and computer-readable storage medium for generating article |
| CN109657069B (en) * | 2018-12-11 | 2021-03-23 | 北京百度网讯科技有限公司 | Knowledge graph generation method and device |
| CN109697050B (en) * | 2019-01-07 | 2021-04-27 | 浙江大学 | Demand description model design method facing E-commerce field based on knowledge graph |
| CN109828965B (en) * | 2019-01-09 | 2021-06-15 | 千城数智(北京)网络科技有限公司 | Data processing method and electronic equipment |
| CN109829116B (en) * | 2019-02-14 | 2021-07-30 | 北京达佳互联信息技术有限公司 | Content recommendation method and device, server and computer readable storage medium |
| CN111782774B (en) * | 2019-04-03 | 2024-04-19 | 北京嘀嘀无限科技发展有限公司 | Method and device for recommending problems |
| CN110555753A (en) * | 2019-08-14 | 2019-12-10 | 中国平安人寿保险股份有限公司 | recommendation-based ranking control method and device, computer equipment and storage medium |
| CN110674394B (en) * | 2019-08-20 | 2024-03-15 | 腾讯科技（深圳）有限公司 | Knowledge graph-based information recommendation method, device and storage medium |
| CN110728783A (en) * | 2019-08-31 | 2020-01-24 | 苏州浪潮智能科技有限公司 | Self-correction method, system and equipment of face recognition system |
| CN110717099B (en) * | 2019-09-25 | 2022-04-22 | 优地网络有限公司 | Method and terminal for recommending film |
| CN110888990B (en) * | 2019-11-22 | 2024-04-12 | 深圳前海微众银行股份有限公司 | Text recommendation method, device, equipment and medium |
| CN111428053B (en) * | 2020-03-30 | 2023-10-20 | 西安交通大学 | Construction method of tax field-oriented knowledge graph |
| CN111949803B (en) * | 2020-08-21 | 2024-05-28 | 深圳供电局有限公司 | Knowledge graph-based network abnormal user detection method, device and equipment |
| CN112182115A (en) * | 2020-09-10 | 2021-01-05 | 珠海格力电器股份有限公司 | Relationship display method and device, storage medium and electronic device |
| CN112291272B (en) * | 2020-12-24 | 2021-05-11 | 鹏城实验室 | Network threat detection method, device, equipment and computer readable storage medium |
| CN112950321A (en) * | 2021-03-10 | 2021-06-11 | 北京汇钧科技有限公司 | Article recommendation method and device |
| CN113536137A (en) * | 2021-08-13 | 2021-10-22 | 北京字节跳动网络技术有限公司 | Information display method and device and computer storage medium |
| CN117787392A (en) * | 2024-02-23 | 2024-03-29 | 支付宝(杭州)信息技术有限公司 | Knowledge graph fusion method and device |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103593792A (en) * | 2013-11-13 | 2014-02-19 | 复旦大学 | Individual recommendation method and system based on Chinese knowledge mapping |
| CN104090963A (en) * | 2014-07-14 | 2014-10-08 | 百度在线网络技术（北京）有限公司 | Search information recommendation method and device |
| CN104346446A (en) * | 2014-10-27 | 2015-02-11 | 百度在线网络技术（北京）有限公司 | Paper associated information recommendation method and device based on mapping knowledge domain |
| CN104484459A (en) * | 2014-12-29 | 2015-04-01 | 北京奇虎科技有限公司 | Method and device for combining entities in knowledge map |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101114294A (en) * | 2007-08-22 | 2008-01-30 | 杭州经合易智控股有限公司 | Self-help intelligent uprightness searching method |
| CN101436191A (en) * | 2007-11-13 | 2009-05-20 | 日电（中国）有限公司 | Method and system for calculating competitiveness measurement between calculation objects |
| CN104035917B (en) * | 2014-06-10 | 2017-07-07 | 复旦大学 | A kind of knowledge mapping management method and system based on semantic space mapping |
| CN104462293A (en) * | 2014-11-27 | 2015-03-25 | 百度在线网络技术（北京）有限公司 | Search processing method and method and device for generating search result ranking model |
| CN105335519B (en) * | 2015-11-18 | 2021-08-17 | 百度在线网络技术（北京）有限公司 | Model generation method and device and recommendation method and device |
-
2015
- 2015-11-18 CN CN201510794561.8A patent/CN105335519B/en active Active
-
2016
- 2016-07-11 WO PCT/CN2016/089648 patent/WO2017084362A1/en active Application Filing
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103593792A (en) * | 2013-11-13 | 2014-02-19 | 复旦大学 | Individual recommendation method and system based on Chinese knowledge mapping |
| CN104090963A (en) * | 2014-07-14 | 2014-10-08 | 百度在线网络技术（北京）有限公司 | Search information recommendation method and device |
| CN104346446A (en) * | 2014-10-27 | 2015-02-11 | 百度在线网络技术（北京）有限公司 | Paper associated information recommendation method and device based on mapping knowledge domain |
| CN104484459A (en) * | 2014-12-29 | 2015-04-01 | 北京奇虎科技有限公司 | Method and device for combining entities in knowledge map |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105335519A (en) | 2016-02-17 |
| WO2017084362A1 (en) | 2017-05-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105335519B (en) | Model generation method and device and recommendation method and device | |
| CN109190043B (en) | Recommendation method and device, storage medium, electronic device and recommendation system | |
| CN109166017B (en) | Push method and device based on re-clustering, computer equipment and storage medium | |
| CN108287864B (en) | Interest group dividing method, device, medium and computing equipment | |
| US10180967B2 (en) | Performing application searches | |
| KR20190038751A (en) | User keyword extraction apparatus, method and computer readable storage medium | |
| US10489511B2 (en) | Content editing using AI-based content modeling | |
| CN108108821A (en) | Model training method and device | |
| EP2778960A1 (en) | Generating insightfull connections between graph entities | |
| CN103258025B (en) | Generate the method for co-occurrence keyword, the method that association search word is provided and system | |
| CN109325182B (en) | Information pushing method and device based on session, computer equipment and storage medium | |
| KR20160057475A (en) | System and method for actively obtaining social data | |
| CN109948036B (en) | Method and device for calculating weight of participle term | |
| CN105809473B (en) | Training method for matching model parameters, service recommendation method and corresponding device | |
| US20150248721A1 (en) | Recommendation engine with profile analysis | |
| US20160125028A1 (en) | Systems and methods for query rewriting | |
| CN109241243B (en) | Candidate document sorting method and device | |
| US10467307B1 (en) | Grouping of item data using seed expansion | |
| EP4064071A1 (en) | Text search method, device, server, and storage medium | |
| CN104536979B (en) | The generation method and device of topic model, the acquisition methods and device of theme distribution | |
| CN105512156B (en) | Click model generation method and device | |
| CN112446727B (en) | Advertisement triggering method, device, equipment and computer readable storage medium | |
| EP2766826B1 (en) | Searching information | |
| WO2018058118A1 (en) | Method, apparatus and client of processing information recommendation | |
| CN104298785A (en) | Searching method for public searching resources |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |