CN110020910B - Object recommendation method and device - Google Patents

Object recommendation method and device Download PDF

Info

Publication number
CN110020910B
CN110020910B CN201910063963.9A CN201910063963A CN110020910B CN 110020910 B CN110020910 B CN 110020910B CN 201910063963 A CN201910063963 A CN 201910063963A CN 110020910 B CN110020910 B CN 110020910B
Authority
CN
China
Prior art keywords
user
feature representation
matrix
generating
representation
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
Application number
CN201910063963.9A
Other languages
Chinese (zh)
Other versions
CN110020910A (en
Inventor
胡斌斌
张志强
周俊
李小龙
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Advanced New Technologies Co Ltd
Advantageous New Technologies Co Ltd
Original Assignee
Advanced New Technologies Co Ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Advanced New Technologies Co Ltd filed Critical Advanced New Technologies Co Ltd
Priority to CN201910063963.9A priority Critical patent/CN110020910B/en
Publication of CN110020910A publication Critical patent/CN110020910A/en
Application granted granted Critical
Publication of CN110020910B publication Critical patent/CN110020910B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

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
    • 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/06—Buying, selling or leasing transactions
    • G06Q30/0601—Electronic shopping [e-shopping]
    • G06Q30/0631—Item recommendations

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Databases & Information Systems (AREA)
  • Business, Economics & Management (AREA)
  • General Engineering & Computer Science (AREA)
  • Computational Linguistics (AREA)
  • Artificial Intelligence (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Accounting & Taxation (AREA)
  • Finance (AREA)
  • Health & Medical Sciences (AREA)
  • Economics (AREA)
  • Marketing (AREA)
  • Strategic Management (AREA)
  • General Business, Economics & Management (AREA)
  • Development Economics (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

The specification discloses an object recommendation method and device. The method comprises the following steps: aiming at any user, generating local feature representation for the user according to the feature representation of an associated object which has a direct association relation with the user; aiming at any object, generating local feature representation for the object according to the feature representation of an associated user having a direct association relation with the object; generating a relation characteristic representation between the user and the object according to the indirect incidence relation between the user and the object; calculating a sum of the local feature representation of the user and the relational feature representation and calculating a difference of the sum and the local feature representation of the object; and if the difference meets a preset recommendation condition, recommending the object to the user.

Description

Object recommendation method and device
Technical Field
The specification relates to the field of artificial intelligence, in particular to an object recommendation method and device.
Background
With the rapid development of internet technology, more and more application scenes are needed for object recommendation. For example, the e-commerce platform may recommend goods for the user, the movie ticket purchasing platform may recommend movies for the user, and the like. The accuracy of the recommendation algorithm will directly affect the user experience.
Disclosure of Invention
In view of the above, the present specification provides an object recommendation method and apparatus.
Specifically, the description is realized by the following technical scheme:
an object recommendation method comprising:
aiming at any user, generating local feature representation for the user according to the feature representation of an associated object which has a direct association relation with the user;
aiming at any object, generating local feature representation for the object according to the feature representation of an associated user having a direct association relation with the object;
generating a relation characteristic representation between the user and the object according to the indirect incidence relation between the user and the object;
calculating a sum of the local feature representation of the user and the relational feature representation and calculating a difference of the sum and the local feature representation of the object;
and if the difference meets a preset recommendation condition, recommending the object to the user.
An object recommendation apparatus comprising:
the user feature generation unit is used for generating local feature representation for any user according to the self feature representation of the associated object which has the direct association relation with the user;
the object feature generation unit is used for generating local feature representation for any object according to the self feature representation of the associated user with the direct association relation with the object;
the relation characteristic generating unit is used for generating relation characteristic representation between the user and the object according to the indirect incidence relation between the user and the object;
a feature difference calculation unit that calculates a sum of the local feature representation of the user and the relational feature representation, and calculates a difference between the sum and the local feature representation of the object;
and the object recommending unit recommends the object to the user if the difference meets a preset recommending condition.
An object recommendation apparatus comprising:
a processor;
a memory for storing machine executable instructions;
wherein, by reading and executing machine-executable instructions stored by the memory that correspond to object recommendation logic, the processor is caused to:
aiming at any user, generating local feature representation for the user according to the feature representation of an associated object which has a direct association relation with the user;
aiming at any object, generating local feature representation for the object according to the feature representation of an associated user having a direct association relation with the object;
generating a relation characteristic representation between the user and the object according to the indirect incidence relation between the user and the object;
calculating a sum of the local feature representation of the user and the relational feature representation and calculating a difference of the sum and the local feature representation of the object;
and if the difference meets a preset recommendation condition, recommending the object to the user.
It can be seen from the above description that the relationship feature in the object recommendation scheme of the present specification carries global information between the user and the object, and can avoid the cold start problem to a certain extent. In addition, the object recommendation scheme fuses local information and global information of the user and the object, effectively utilizes various auxiliary information, and can greatly improve the hit rate of object recommendation and the accuracy of a recommendation list.
Drawings
Fig. 1 is a flowchart illustrating an object recommendation method according to an exemplary embodiment of the present disclosure.
Fig. 2 is a flowchart illustrating another object recommendation method according to an exemplary embodiment of the present disclosure.
Fig. 3 is a flowchart illustrating a method for generating a local feature representation according to an exemplary embodiment of the present disclosure.
Fig. 4 is a flowchart illustrating a method for generating a relationship feature representation between a user and an object according to an exemplary embodiment of the present specification.
Fig. 5 is a schematic structural diagram of an object recommending apparatus according to an exemplary embodiment of the present specification.
Fig. 6 is a block diagram of an object recommending apparatus according to an exemplary embodiment of the present specification.
Detailed Description
Reference will now be made in detail to the exemplary embodiments, examples of which are illustrated in the accompanying drawings. When the following description refers to the accompanying drawings, like numbers in different drawings represent the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present specification. Rather, they are merely examples of apparatus and methods consistent with certain aspects of the specification, as detailed in the appended claims.
The terminology used in the description herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the description. As used in this specification 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 also be understood that the term "and/or" as used herein refers to and encompasses any and all possible combinations of one or more of the associated listed items.
It should be understood that although the terms first, second, third, etc. may be used herein to describe various information, these information should not be limited to these terms. These terms are only used to distinguish one type of information from another. For example, the first information may also be referred to as second information, and similarly, the second information may also be referred to as first information, without departing from the scope of the present specification. The word "if" as used herein may be interpreted as "at … …" or "when … …" or "in response to a determination", depending on the context.
The present specification provides an object recommendation scheme.
In one aspect, a local feature representation may be generated for a user based on a self-feature representation of an associated object that has a direct association with the user, and a local feature representation may be generated for an object based on a self-feature representation of an associated user that has a direct association with the object.
On the other hand, the relation characteristic representation between the user and the object carrying the global information can be generated according to the indirect incidence relation between the user and the object.
Then, a sum of the user local feature representation and the above-described relational feature representation and a difference between the sum and the object local feature representation may be calculated, and the object may be recommended to the user when the difference satisfies a predetermined recommendation condition.
The relationship characteristic representation in the object recommendation scheme carries global information between the user and the object, and the cold start problem can be avoided to a certain extent. In addition, the object recommendation scheme fuses local information and global information of the user and the object, effectively utilizes various auxiliary information, and can greatly improve the hit rate of object recommendation and the accuracy of a recommendation list.
The above objects may include: goods or services. For example, the e-commerce platform may recommend goods or services for sale to the user.
The above object may further include: information is obtained. For example, a web portal may recommend information to a user.
The above object may further include: and (6) video. For example, a video APP (Application) may recommend videos such as a tv show, a movie, and the like to a user.
Of course, in other examples, the object may also be other entities or data, and the object recommendation may also be applied in other application scenarios, which is not particularly limited in this specification.
The implementation of the present specification is described below with reference to specific embodiments.
Referring to fig. 1 and 2, the object recommendation method provided in the present specification may include the following steps:
step 102, aiming at any user, generating local feature representation for the user according to the feature representation of the associated object which has direct association relation with the user.
And 104, aiming at any object, generating local feature representation for the object according to the self feature representation of the associated user having the direct association relation with the object.
In this embodiment, the direct association relationship may include: purchase, browse, review, etc.
Taking the object as an example, the items directly related to the user may include: items purchased by the user, items viewed by the user, items reviewed by the user, items attended by the user, and the like. Generally, there are a plurality of commodities that are directly related to the same user.
Users having a direct association with the merchandise may include: users who have purchased the good, users who have browsed the good, users who have reviewed the good, users who have attended to the good, and the like. Similarly, there are usually a plurality of users who have direct association with the same product.
Taking the example that the object is a movie, movies having a direct association with the user may include: movies viewed by the user, movies rated by the user, movies collected by the user, etc.
Users having a direct association with a movie may include: users who have viewed the movie, users who have commented on the movie, users who have collected the movie, etc.
Referring to fig. 3, the process of generating the local feature representation of the user and the object may include the following steps:
step 302, generating a related object feature representation matrix according to the feature representation of each related object.
And 304, generating a related user feature representation matrix according to the feature representation of each related user.
In this embodiment, the self-feature representation of each object may be generated first, for example, a corresponding 0/1 vector may be generated for each object, and then the 0/1 vector may be embedded to obtain the self-feature representation of each object.
The 0/1 vector is a very long vector used to represent the object, the dimension of the vector is the total number of objects, the element value of the dimension of the object is 1, and the element values of the other dimensions are all 0.
Assuming that there are a total of 1000 ten thousand objects, the 0/1 vector of each object is 1000 ten thousand dimensions, the element value of the 1 st dimension of object 1 is 1, and the element values of the other dimensions are 0, then the 0/1 vector of object 1 can be represented as [1,0,0,0 … ]; the element value of the 3 rd dimension of object 2 is 1 and the other dimension element values are 0, then the 0/1 vector for object 2 can be represented as [0,0,1,0,0 … ].
In this embodiment, an Embedding process (Embedding) may be performed on the 0/1 vector of each object, and the 0/1 vector of the high dimension is mapped to the low dimension space, so as to obtain the low dimension feature representation of each object. Since the low-dimensional feature representation does not include heterogeneous information such as attributes of objects and relationships between objects, the low-dimensional feature representation can be referred to as a self-feature representation of the object.
In this embodiment, the self-feature representations of all the objects may be generated in advance, and then after the associated object of a certain user is determined, the self-feature representation of each associated object is acquired.
Similarly, the self-feature representations of all users may be generated in advance, and then the self-feature representation of each associated user may be acquired after the associated user of a certain object is determined. The generation process of the user self feature representation may refer to the generation process of the object self feature representation, and this description is not repeated here.
In this embodiment, the self-characterization of all the associated objects of each user may constitute an associated object characterization matrix.
For example, assuming that the feature representation of the associated object is a D-dimensional vector, and a user has M associated objects, an M × D associated object feature representation matrix may be formed.
The self feature representation of all the associated users of each object can also form an associated user feature representation matrix.
For example, assuming that the feature representation of the associated user is also a D-dimensional vector, and there are N associated users in a certain object, an N × D associated user feature representation matrix may be formed.
Step 306, generating a cooperation matrix of the associated object and the associated user according to the associated object feature representation matrix and the associated user feature representation matrix.
In this embodiment, still taking an M × D associated object feature representation matrix and an N × D associated user feature representation matrix as an example, a product of the associated object feature representation matrix and the associated user feature representation matrix may be calculated to obtain an M × N cooperation matrix. The collaborative matrix integrates the self-feature representations of the associated objects and the associated users.
In other examples, the trained neural network model may be used to perform nonlinear transformation on the associated object feature representation matrix and the associated user feature representation matrix, and then the product of the transformed matrices is calculated to obtain the cooperation matrix.
Specifically, the M × D associated object feature representation matrix may be input into the trained first neural network model, for example, an M × D 'matrix may be output, and for convenience of description, the M × D' matrix may be referred to as a transformed associated object matrix. The first neural network model may be a model such as a multilayer perceptron, and the specification does not limit this.
Similarly, the N × D associated user feature representation matrix may be input into the trained second neural network model, for example, an N × D 'matrix may be output, and for convenience of description, the N × D' matrix may be referred to as a transformed associated user matrix.
Then, the product of the M × D 'transformation correlation object matrix and the N × D' transformation correlation user matrix can be calculated to obtain the M × N cooperation matrix.
The neural network model is adopted to transform the associated object feature representation matrix and the associated user feature representation matrix, so that the fitting capability of the model can be enhanced.
Of course, in other examples, the collaborative matrix of the associated object and the associated user may also be generated in other manners, which is not particularly limited in this specification.
And 308, performing row pooling and column pooling on the collaborative matrix respectively to obtain the attention vectors of the associated object and the associated user.
In this embodiment, the pooling process may include: maximum pooling, average pooling, minimum pooling, and the like, which are not particularly limited in this specification.
Taking the maximum pooling as an example, on one hand, the maximum pooling processing may be performed on each row of the M × N cooperative matrix, that is, for each row of the cooperative matrix, the maximum element value is selected as the pooling result of the row to obtain the attention vector of the associated object, where the attention vector of the associated object includes M elements.
On the other hand, each column of the M × N collaborative matrix may be maximally pooled, that is, for each column of the collaborative matrix, a maximum element value is selected as a pooling result of the column to obtain an attention vector of an associated user, where the attention vector of the associated user includes N elements.
In the present embodiment, after the pooling process is performed, the attention vector may be further normalized by using a softmax function so that each element value of the attention vector is between 0 and 1 and the sum of the element values is 1.
Step 310, generating a local feature representation for the user based on the associated object feature representation matrix and the attention vector of the associated object.
In this embodiment, each element value in the associated object attention vector may be used as a weight value, and elements from different associated objects but with the same dimension in the associated object feature representation matrix may be weighted and summed, respectively, to generate a feature representation for the user.
Assume that the M × D associated object feature representation matrix is as follows:
Figure BDA0001955077980000081
the associated object attention vector is: (w)1，w2，…，wM) Then, the first element value of the user local feature representation generated according to the M × D associated object feature representation matrix and the associated object attention vector is:
w1×a11+w2×a12+w3×a13+…+wM×a1M，
the second element value is: w is a1×a21+w2×a22+w3×a23+…+wM×a2M，
Similarly, the Dth element value is:
w1×aD1+w2×aD2+w3×aD3+…+wM×aDM。
step 312, generating a local feature representation for the object based on the associated user feature representation matrix and the attention vector of the associated user.
Similar to the local feature representation of the user, each element value in the attention vector of the associated user may be used as a weight value, and elements from different associated users but with the same dimension in the associated user feature representation matrix may be weighted and summed respectively to generate a local feature representation for the object.
In this embodiment, the attention vectors of the associated objects and the associated users are obtained through pooling processing of the collaborative matrix, and then local feature representations of the users and the objects are generated based on the attention vectors, so that the influence degrees of different associated objects and different associated users on the current recommendation can be distinguished, and the accuracy of subsequent object recommendation is further improved.
And 106, generating a relation characteristic representation between the user and the object according to the indirect incidence relation between the user and the object.
In this embodiment, the indirect association relationship may mean that there is no direct association relationship between the user and the object, but the user and the object may be associated together through other users and objects.
For example, the third Zhang of the user does not watch the movie bumblebee, and there is no other direct association relationship such as concern and comment between the third Zhang of the user and the bumblebee, but the friend of the third Zhang of the user is plum and watches the bumblebee, and in this case, there is an indirect association relationship between the third Zhang of the user and the bumblebee.
For another example, the user has not purchased the iPhone XS, and there is no direct association between the user's whites and the iPhone XS, such as attention, comment, browsing, etc., but the friend of the whites has purchased the iPhone8, in which case the whites are associated with the iPhone XS through the whites, the iPhone8, the apple brand name, and the iPhone XS, and there is also an indirect association between the whites and the iPhone XS.
In this embodiment, referring to fig. 4, the process of generating the relationship feature representation between the user and the object may include the following steps:
step 402, obtaining a plurality of indirectly reachable paths between the user and the object, where the indirectly reachable paths connect the user and the object through one or more nodes, where a connecting edge on the indirectly reachable path represents that there is a direct association relationship between the connected nodes, and the types of the nodes include: user class nodes, object class nodes, and non-user and non-object class nodes.
In this embodiment, a heterogeneous network carrying various information may be constructed first.
The heterogeneous network may include user class nodes, object class nodes, and non-user and non-object class nodes. For example, a heterogeneous network may be constructed according to information such as user attributes, object attributes, association between users and objects, association between users and users, and the like, and non-user and non-object class nodes in the heterogeneous network may represent user attributes, object attributes, and the like.
When a connecting edge of the heterogeneous network connects an object and a user, the connecting edge can represent that a direct association relationship exists between the connected object and the user.
When a connection edge of the heterogeneous network connects a user and a user attribute, the connection edge may represent that the user has the connected user attribute.
When the connecting edge of the heterogeneous network connects the user and the user, the connecting edge can represent that the connected users have the association relationship of friends, transfer accounts and the like.
Certainly, the connection edge in the heterogeneous network may also connect the user and the object attribute, and the like, which is not described herein any more.
After the heterogeneous network is constructed, a specified number of nodes can be randomly selected from the heterogeneous network, and then path extraction is performed by adopting a random walk mode from the selected nodes.
Taking the specified number as 1000 as an example, 1000 nodes may be randomly selected from the heterogeneous network. For each selected node, a path with a specified hop count, such as 100 hops, may be extracted from the node in a random walk manner.
In this embodiment, 1000 paths may be extracted from the heterogeneous network, and the number of each path is 100.
Of course, in other examples, the number of each path may not be exactly the same, and the present specification does not particularly limit this.
In this embodiment, by setting the number of nodes and the hop count of the path, it can be considered that the extracted path carries most information of the heterogeneous network.
In this embodiment, whether the extracted paths include an indirect reachable path between the user and the object may be separately determined, and if the extracted paths include an indirect reachable path, the indirect reachable paths may be extracted.
For example, assuming that the user node is u2 and the object node is m5, 4 paths are extracted from the heterogeneous network, which are:
route 1: u1-m1-u2-m3-t1-m4-u3-m5-d1-m 8;
route 2: u8-m1-u2-m3-a7-m 5;
route 3: u5-m1-u2-m3-t1-m4-m 5;
path 4: u2-m3-t1-m 4.
Wherein, the path 1 includes an indirect reachable path between the user u2 and the object m 5: u2-m3-t1-m4-u3-m 5;
also included in path 2 is an indirectly reachable path between user u2 to object m 5: u2-m3-a7-m 5;
also included in path 3 is an indirectly reachable path between user u2 to object m 5: u2-m3-t1-m4-m 5;
the indirect reachable path between user u2 to object m5 is not included in path 4.
In this example, 3 indirectly reachable paths between user u2 to object m5 may be extracted.
Of course, in other examples, other manners may also be used to extract the indirect reachable path, which is not limited in this specification.
And step 404, aiming at each indirect reachable path, generating the characteristic representation of the indirect reachable path according to the characteristic representation of each node on the indirect reachable path.
Taking the aforementioned indirect reachable path u2-m3-t1-m4-u3-m5 as an example, in this embodiment, the self feature representation of each node (nodes u2, m3, t1, m4, u3, and m5) in the indirect reachable path may be obtained first, and then the feature representation of the indirect reachable path is generated by synthesizing the self feature representations of each node.
For example, the feature representation of each node on the indirect reachable path may be input, and the feature representation of the indirect reachable path may be output by using a trained LSTM (Long Short-Term Memory) model.
In this embodiment, the method for generating the self-feature representation of the non-user and non-object class nodes in the indirect reachable path may refer to the method for generating the self-feature representation of the user class node and the object class node, which is not described in detail herein.
And 406, fusing the feature representation of each indirect reachable path to obtain a relationship feature representation between the user and the object.
Based on the foregoing step 404, after generating the feature representation for each indirectly reachable path, the feature representations of all indirectly reachable paths between the user and the object may be fused to obtain the relationship feature representation between the user and the object.
In one example, the feature representations of the indirectly reachable paths may be fused by summing, averaging, and the like. For example, the relational feature representation is obtained by performing operations such as summing and averaging on the element values of the same dimension in the feature representation of each indirect reachable path.
In another example, the weight value of each indirectly reachable path may be determined according to the hop count of the indirectly reachable path, and then the feature representations of the indirectly reachable paths may be fused by means of weighted summation or weighted average.
Generally, the smaller the hop count of the indirect reachable path is, the greater the influence of the path on the recommendation is, and therefore the weight is higher; conversely, the more the number of hops of the indirect reachable path is, the smaller the influence of the path on the recommendation is, and therefore the weight is lower.
Indirectly reachable path Hop count Weighted value
u2-m3-t1-m4-u3-m5 5 3/12
u2-m3-a7-m5 3 5/12
u2-m3-t1-m4-m5 4 4/12
TABLE 1
Referring to table 1, still taking the aforementioned 3 indirect reachable paths from the user u2 to the object m5 as an example, the hop count and the weight value of each indirect reachable path may refer to the example of table 1. The weight value of the indirect reachable path u2-m3-t1-m4-u3-m5 with the largest hop count is the smallest, and the weight value of the indirect reachable path u2-m3-a7-m5 with the smallest hop count is the largest.
In this example, each element value in the feature representation of each indirect reachable path may be multiplied by the weight value of the indirect reachable path, and then the element values of the same dimension are summed and averaged, so as to obtain the relationship feature representation between the user and the object.
Of course, in other examples, the feature representations of the indirect reachable paths may be fused in other ways, and this specification does not limit this.
Step 108, calculating a sum of the local feature representation of the user and the relational feature representation, and calculating a difference between the sum and the local feature representation of the object.
In this embodiment, when the local feature representation and the relational feature representation are vectors, a vector sum of a local vector of the user and a relational vector may be calculated, and then a vector difference between the vector sum and the target local vector may be calculated as the difference.
When the local feature representation and the relational feature representation are matrices, a matrix sum of a local matrix of the user and the relational matrix may be calculated, and then a matrix difference between the matrix sum and the object local matrix may be calculated as the difference.
Similarly, when the feature representation is in other forms, the sum and the difference can be calculated in a corresponding manner, and the description of the present specification is omitted.
And step 110, recommending the object to the user if the difference meets a preset recommendation condition.
In this embodiment, the recommendation condition may be determined in a training phase, for example, the recommendation condition may be that the difference is smaller than a difference threshold.
In one example, when the discrepancy is a vector, a modulus of the vector may be calculated and then a determination may be made as to whether the modulus of the vector is less than a discrepancy threshold.
In this embodiment, when the difference satisfies the recommendation condition, it may be described that the sum of the local feature representation of the user and the relationship feature representation between the user and the object is closer to the local feature representation of the object, and the object may be recommended to the user.
As can be seen from the above description, the relationship characteristic representation in the object recommendation scheme provided in this specification carries global information between the user and the object, and can avoid the cold start problem to a certain extent. In addition, the object recommendation scheme fuses local information and global information of the user and the object, effectively utilizes various auxiliary information, and can greatly improve the hit rate of object recommendation and the accuracy of a recommendation list.
In the foregoing embodiment shown in fig. 1, the parameters of the first neural network model, the second neural network model, the LSTM model for generating the indirect reachable path feature representation, and the like, and the recommended conditions may be determined at the time of training.
In this embodiment, the model and the recommendation condition may be trained by the method shown in fig. 1 based on the direct association relationship between the user and the object that has occurred historically as a positive sample. For example, if the user's floret watched the movie spy on disc 6, the user's floret and movie spy on disc 6 could be trained as a positive sample.
Optionally, in order to accelerate the training, a negative sample can be added in the training process. For example, if the user floret and the movie spy in disc 5 do not have any direct relationship, the user floret and spy in disc 5 may be trained as a negative example.
In the training, a loss function may be constructed based on the difference calculated in the step 108, for example, a hinge loss function loss may be constructed as max(s)Is just for-sNegative pole+ m,0), wherein sIs justDenotes the difference of the positive samples, sNegative poleRepresenting the difference of negative examples, m may represent a difference threshold in the recommended conditions.
Of course, in other examples, training may be performed in other manners, and the present specification is not limited thereto.
Corresponding to the embodiments of the object recommendation method, the present specification also provides embodiments of an object recommendation device.
The embodiment of the object recommendation device can be applied to a server. The device embodiments may be implemented by software, or by hardware, or by a combination of hardware and software. Taking a software implementation as an example, as a logical device, the device is formed by reading corresponding computer program instructions in the nonvolatile memory into the memory for operation through the processor of the server where the device is located. From a hardware aspect, as shown in fig. 5, the hardware structure diagram of the server where the object recommendation device is located in this specification is shown, except for the processor, the memory, the network interface, and the nonvolatile memory shown in fig. 5, the server where the device is located in the embodiment may also include other hardware according to the actual function of the server, which is not described again.
Fig. 6 is a block diagram of an object recommending apparatus according to an exemplary embodiment of the present specification.
Referring to fig. 6, the object recommendation apparatus 500 may be applied to the server shown in fig. 5, and includes: a user feature generation unit 501, an object feature generation unit 502, a relationship feature generation unit 503, a feature difference calculation unit 504, and an object recommendation unit 505.
The user feature generation unit 501 generates, for any user, a local feature representation for the user according to a feature representation of an associated object having a direct association relationship with the user;
an object feature generation unit 502 that generates, for any object, a local feature representation for the object from a feature representation of an associated user who has a direct association with the object;
a relation feature generation unit 503, configured to generate a relation feature representation between the user and the object according to an indirect association relation between the user and the object;
a feature difference calculation unit 504 that calculates a sum of the local feature representation of the user and the relational feature representation, and calculates a difference between the sum and the local feature representation of the object;
and an object recommending unit 505 that recommends the object to the user if the difference satisfies a predetermined recommending condition.
Optionally, the process of generating the local feature representation of the user and the object includes:
generating a related object feature representation matrix according to the feature representation of each related object;
generating a correlation user feature representation matrix according to the feature representation of each correlation user;
generating a cooperative matrix of the associated object and the associated user according to the associated object feature representation matrix and the associated user feature representation matrix;
respectively performing row pooling and column pooling on the collaborative matrix to obtain the attention vectors of the associated object and the associated user;
generating a local feature representation for the user based on the associated object feature representation matrix and the attention vector of the associated object;
generating a local feature representation for the object based on the matrix of associated user feature representations and the attention vector of the associated user.
Optionally, the generating a cooperation matrix of the associated object and the associated user according to the associated object feature representation matrix and the associated user feature representation matrix includes:
taking the correlation object feature representation matrix as input, and outputting a transformation correlation object matrix by adopting a trained first neural network model;
taking the associated user feature representation matrix as input, and outputting a transformation associated user matrix by adopting a trained second neural network model;
and calculating the product of the transformation associated object matrix and the transformation associated user matrix to obtain the cooperative matrix.
Optionally, the generating a local feature representation for the user based on the associated object feature representation matrix and the attention vector of the associated object includes:
taking each element value in the associated object attention vector as a weight value, and respectively carrying out weighted summation on elements which come from different associated objects and have the same dimension in the associated object feature representation matrix so as to generate local feature representation for the user;
the generating a local feature representation for the object based on the associated user feature representation matrix and the attention vector of the associated user comprises:
and taking each element value in the associated user attention vector as a weight value, and respectively carrying out weighted summation on elements which come from different associated users and have the same dimension in the associated user feature representation matrix so as to generate local feature representation for the object.
Optionally, the relationship characteristic generating unit 503:
acquiring a plurality of indirectly reachable paths between the user and the object, wherein the indirectly reachable paths connect the user and the object through one or more nodes, connecting edges on the indirectly reachable paths represent that the connected nodes have a direct association relationship, and the types of the nodes include: user class nodes, object class nodes and non-user and non-object class nodes;
aiming at each indirect reachable path, generating the characteristic representation of the indirect reachable path according to the characteristic representation of each node on the indirect reachable path;
and fusing the characteristic representation of each indirect reachable path to obtain the relation characteristic representation between the user and the object.
Optionally, the obtaining a plurality of indirectly reachable paths between the user and the object includes:
constructing a heterogeneous network comprising user class nodes, object class nodes and non-user and non-object class nodes, wherein connecting edges in the heterogeneous network represent that the connected nodes have a direct association relationship;
randomly selecting a specified number of nodes from the heterogeneous network, and respectively starting from the selected nodes to extract paths in a random walk mode;
and acquiring an indirect reachable path between the user and the object from the extracted path.
Optionally, the generating a feature representation of the indirect reachable path according to the feature representation of each node on the indirect reachable path includes:
and taking the characteristic representation of each node on the indirect reachable path as input, and outputting the characteristic representation of the indirect reachable path by adopting an LSTM model.
Optionally, the fusing the feature representation of each indirect reachable path to obtain a relationship feature representation between the user and the object includes:
determining a weight value of the indirect reachable path according to the hop count of the indirect reachable path;
and according to the weight value, fusing the feature representation of each indirect reachable path in a weighted average mode to obtain the relationship feature representation between the user and the object.
Optionally, the predetermined recommendation condition includes: the difference is less than or equal to a difference threshold.
The implementation process of the functions and actions of each unit in the above device is specifically described in the implementation process of the corresponding step in the above method, and is not described herein again.
For the device embodiments, since they substantially correspond to the method embodiments, reference may be made to the partial description of the method embodiments for relevant points. The above-described embodiments of the apparatus are merely illustrative, and 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 modules can be selected according to actual needs to achieve the purpose of the solution in the specification. One of ordinary skill in the art can understand and implement it without inventive effort.
The systems, devices, modules or units illustrated in the above embodiments may be implemented by a computer chip or an entity, or by a product with certain functions. A typical implementation device is a computer, which may take the form of a personal computer, laptop computer, cellular telephone, camera phone, smart phone, personal digital assistant, media player, navigation device, email messaging device, game console, tablet computer, wearable device, or a combination of any of these devices.
Corresponding to the foregoing embodiments of the object recommendation method, the present specification also provides an object recommendation apparatus, including: a processor and a memory for storing machine executable instructions. Wherein the processor and the memory are typically interconnected by means of an internal bus. In other possible implementations, the device may also include an external interface to enable communication with other devices or components.
In this embodiment, the processor is caused to:
aiming at any user, generating local feature representation for the user according to the feature representation of an associated object which has a direct association relation with the user;
aiming at any object, generating local feature representation for the object according to the feature representation of an associated user having a direct association relation with the object;
generating a relation characteristic representation between the user and the object according to the indirect incidence relation between the user and the object;
calculating a sum of the local feature representation of the user and the relational feature representation and calculating a difference of the sum and the local feature representation of the object;
and if the difference meets a preset recommendation condition, recommending the object to the user.
Optionally, upon generation of the local feature representation of the user and the object, the processor is caused to:
generating a related object feature representation matrix according to the feature representation of each related object;
generating a correlation user characteristic representation matrix according to the self characteristic representation of each correlation user;
generating a cooperative matrix of the associated object and the associated user according to the associated object feature representation matrix and the associated user feature representation matrix;
respectively performing row pooling and column pooling on the collaborative matrix to obtain the attention vectors of the associated object and the associated user;
generating a local feature representation for the user based on the associated object feature representation matrix and the attention vector of the associated object;
generating a local feature representation for the object based on the matrix of associated user feature representations and the attention vector of the associated user.
Optionally, when generating the collaborative matrix of the associated object and the associated user according to the associated object feature representation matrix and the associated user feature representation matrix, the processor is caused to:
taking the correlation object feature representation matrix as an input, and outputting a transformation correlation object matrix by adopting a trained first neural network model;
taking the associated user feature representation matrix as an input, and outputting a transformation associated user matrix by adopting a trained second neural network model;
and calculating the product of the transformation associated object matrix and the transformation associated user matrix to obtain the cooperative matrix.
Optionally, when generating the local feature representation for the user based on the associated object feature representation matrix and the attention vector of the associated object, the processor is caused to:
taking each element value in the associated object attention vector as a weight value, and respectively carrying out weighted summation on elements which come from different associated objects and have the same dimension in the associated object feature representation matrix so as to generate local feature representation for the user;
the generating a local feature representation for the object based on the associated user feature representation matrix and the attention vector of the associated user comprises:
and taking each element value in the associated user attention vector as a weight value, and respectively carrying out weighted summation on elements which come from different associated users and have the same dimension in the associated user feature representation matrix so as to generate local feature representation for the object.
Optionally, when generating the relationship feature representation between the user and the object according to the indirect association relationship between the user and the object, the processor is caused to:
acquiring a plurality of indirectly reachable paths between the user and the object, wherein the indirectly reachable paths connect the user and the object through one or more nodes, connecting edges on the indirectly reachable paths represent that the connected nodes have a direct association relationship, and the types of the nodes include: user class nodes, object class nodes and non-user and non-object class nodes;
aiming at each indirect reachable path, generating the characteristic representation of the indirect reachable path according to the characteristic representation of each node on the indirect reachable path;
and fusing the characteristic representation of each indirect reachable path to obtain the relation characteristic representation between the user and the object.
Optionally, in obtaining several indirectly reachable paths between the user and the object, the processor is caused to:
constructing a heterogeneous network comprising user class nodes, object class nodes and non-user and non-object class nodes, wherein connecting edges in the heterogeneous network represent that the connected nodes have a direct association relationship;
randomly selecting a specified number of nodes from the heterogeneous network, and respectively starting from the selected nodes to extract paths in a random walk mode;
and acquiring an indirect reachable path between the user and the object from the extracted path.
Optionally, when generating the feature representation of the indirect reachable path according to the feature representation of each node on the indirect reachable path, the processor is caused to:
and taking the characteristic representation of each node on the indirect reachable path as input, and outputting the characteristic representation of the indirect reachable path by adopting an LSTM model.
Optionally, when fusing the feature representation of each indirectly reachable path to obtain the relationship feature representation between the user and the object, the processor is caused to:
determining a weight value of the indirect reachable path according to the hop count of the indirect reachable path;
and according to the weight value, fusing the feature representation of each indirect reachable path in a weighted average mode to obtain the relationship feature representation between the user and the object.
Optionally, the predetermined recommendation condition includes: the difference is less than or equal to a difference threshold.
In correspondence with the foregoing embodiments of the object recommending method, the present specification also provides a computer-readable storage medium having stored thereon a computer program which, when executed by a processor, realizes the steps of:
aiming at any user, generating local feature representation for the user according to the feature representation of an associated object which has a direct association relation with the user;
aiming at any object, generating local feature representation for the object according to the feature representation of an associated user having a direct association relation with the object;
generating a relation characteristic representation between the user and the object according to the indirect incidence relation between the user and the object;
calculating a sum of the local feature representation of the user and the relational feature representation and calculating a difference of the sum and the local feature representation of the object;
and if the difference meets a preset recommendation condition, recommending the object to the user.
Optionally, the process of generating the local feature representation of the user and the object includes:
generating a related object feature representation matrix according to the feature representation of each related object;
generating a correlation user characteristic representation matrix according to the self characteristic representation of each correlation user;
generating a cooperative matrix of the associated object and the associated user according to the associated object feature representation matrix and the associated user feature representation matrix;
respectively performing row pooling and column pooling on the collaborative matrix to obtain the attention vectors of the associated object and the associated user;
generating a local feature representation for the user based on the associated object feature representation matrix and the attention vector of the associated object;
generating a local feature representation for the object based on the matrix of associated user feature representations and the attention vector of the associated user.
Optionally, the generating a cooperation matrix of the associated object and the associated user according to the associated object feature representation matrix and the associated user feature representation matrix includes:
taking the correlation object feature representation matrix as input, and outputting a transformation correlation object matrix by adopting a trained first neural network model;
taking the associated user feature representation matrix as input, and outputting a transformation associated user matrix by adopting a trained second neural network model;
and calculating the product of the transformation associated object matrix and the transformation associated user matrix to obtain the cooperative matrix.
Optionally, the generating a local feature representation for the user based on the associated object feature representation matrix and the attention vector of the associated object includes:
taking each element value in the associated object attention vector as a weight value, and respectively carrying out weighted summation on elements which come from different associated objects and have the same dimension in the associated object feature representation matrix so as to generate local feature representation for the user;
the generating a local feature representation for the object based on the associated user feature representation matrix and the attention vector of the associated user comprises:
and taking each element value in the associated user attention vector as a weight value, and respectively carrying out weighted summation on elements which come from different associated users and have the same dimension in the associated user feature representation matrix so as to generate local feature representation for the object.
Optionally, the generating a relationship feature representation between the user and the object according to the indirect association relationship between the user and the object includes:
acquiring a plurality of indirectly reachable paths between the user and the object, wherein the indirectly reachable paths connect the user and the object through one or more nodes, connecting edges on the indirectly reachable paths represent that the connected nodes have a direct association relationship, and the types of the nodes include: user class nodes, object class nodes and non-user and non-object class nodes;
aiming at each indirect reachable path, generating the characteristic representation of the indirect reachable path according to the characteristic representation of each node on the indirect reachable path;
and fusing the characteristic representation of each indirect reachable path to obtain the relation characteristic representation between the user and the object.
Optionally, the obtaining a plurality of indirectly reachable paths between the user and the object includes:
constructing a heterogeneous network comprising user class nodes, object class nodes and non-user and non-object class nodes, wherein connecting edges in the heterogeneous network represent that the connected nodes have a direct association relationship;
randomly selecting a specified number of nodes from the heterogeneous network, and respectively starting from the selected nodes to extract paths in a random walk mode;
and acquiring an indirect reachable path between the user and the object from the extracted path.
Optionally, the generating a feature representation of the indirect reachable path according to the feature representation of each node on the indirect reachable path includes:
and taking the characteristic representation of each node on the indirect reachable path as input, and outputting the characteristic representation of the indirect reachable path by adopting an LSTM model.
Optionally, the fusing the feature representation of each indirect reachable path to obtain a relationship feature representation between the user and the object includes:
determining a weight value of the indirect reachable path according to the hop count of the indirect reachable path;
and according to the weight value, fusing the feature representation of each indirect reachable path in a weighted average mode to obtain the relationship feature representation between the user and the object.
Optionally, the predetermined recommendation condition includes: the difference is less than or equal to a difference threshold.
The foregoing description has been directed to specific embodiments of this disclosure. Other embodiments are within the scope of the following claims. In some cases, the actions or steps recited in the claims can be performed in a different order than in the embodiments and still achieve desirable results. In addition, the processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order, to achieve desirable results. In some embodiments, multitasking and parallel processing may also be possible or may be advantageous.
The above description is only a preferred embodiment of the present disclosure, and should not be taken as limiting the present disclosure, and any modifications, equivalents, improvements and the like made within the spirit and principle of the present disclosure should be included in the protection scope of the present disclosure.

Claims (17)

1. An object recommendation method comprising:
aiming at any user, generating local feature representation for the user according to the feature representation of an associated object which has a direct association relation with the user;
aiming at any object, generating local feature representation for the object according to the feature representation of an associated user having a direct association relation with the object;
generating a relation characteristic representation between the user and the object according to the indirect incidence relation between the user and the object;
calculating a sum of the local feature representation of the user and the relational feature representation and calculating a difference of the sum and the local feature representation of the object;
if the difference meets a preset recommendation condition, recommending the object to the user;
wherein the process of generating the local feature representation of the user and the object comprises:
generating a related object feature representation matrix according to the feature representation of each related object;
generating a correlation user characteristic representation matrix according to the self characteristic representation of each correlation user;
generating a cooperative matrix of the associated object and the associated user according to the associated object feature representation matrix and the associated user feature representation matrix;
respectively performing row pooling and column pooling on the cooperative matrix to obtain attention vectors of the associated object and the associated user;
generating a local feature representation for the user based on the associated object feature representation matrix and the attention vector of the associated object;
generating a local feature representation for the object based on the matrix of associated user feature representations and the attention vector of the associated user.
2. The method of claim 1, the generating a co-matrix of the associated objects and the associated users from the associated object feature representation matrix and the associated user feature representation matrix, comprising:
taking the correlation object feature representation matrix as input, and outputting a transformation correlation object matrix by adopting a trained first neural network model;
taking the associated user feature representation matrix as input, and outputting a transformation associated user matrix by adopting a trained second neural network model;
and calculating the product of the transformation associated object matrix and the transformation associated user matrix to obtain the cooperative matrix.
3. The method of claim 1, the generating a local feature representation for the user based on the associated object feature representation matrix and the attention vector of the associated object, comprising:
taking each element value in the associated object attention vector as a weight value, and respectively carrying out weighted summation on elements which come from different associated objects and have the same dimension in the associated object feature representation matrix so as to generate local feature representation for the user;
the generating a local feature representation for the object based on the associated user feature representation matrix and the attention vector of the associated user comprises:
and taking each element value in the associated user attention vector as a weight value, and respectively carrying out weighted summation on elements which come from different associated users and have the same dimension in the associated user feature representation matrix so as to generate local feature representation for the object.
4. The method of claim 1, the generating a relational feature representation between the user and the object according to the indirect associative relationship between the user and the object, comprising:
acquiring a plurality of indirectly reachable paths between the user and the object, wherein the indirectly reachable paths connect the user and the object through one or more nodes, connecting edges on the indirectly reachable paths represent that the connected nodes have a direct association relationship, and the types of the nodes include: user class nodes, object class nodes and non-user and non-object class nodes;
aiming at each indirect reachable path, generating the characteristic representation of the indirect reachable path according to the characteristic representation of each node on the indirect reachable path;
and fusing the characteristic representation of each indirect reachable path to obtain the relation characteristic representation between the user and the object.
5. The method of claim 4, the obtaining a number of indirectly reachable paths between the user and the object, comprising:
constructing a heterogeneous network comprising user class nodes, object class nodes and non-user and non-object class nodes, wherein connecting edges in the heterogeneous network represent that the connected nodes have a direct association relationship;
randomly selecting a specified number of nodes from the heterogeneous network, and respectively starting from the selected nodes to extract paths in a random walk mode;
and acquiring an indirect reachable path between the user and the object from the extracted path.
6. The method according to claim 4, wherein said generating the feature representation of the indirectly reachable path according to the feature representation of each node on the indirectly reachable path comprises:
and taking the characteristic representation of each node on the indirect reachable path as input, and outputting the characteristic representation of the indirect reachable path by adopting an LSTM model.
7. The method of claim 4, wherein fusing the feature representation of each indirectly reachable path to obtain a relationship feature representation between the user and the object comprises:
determining a weight value of the indirect reachable path according to the hop count of the indirect reachable path;
and according to the weight value, fusing the feature representation of each indirect reachable path in a weighted average mode to obtain the relationship feature representation between the user and the object.
8. The method of claim 1, wherein the first and second light sources are selected from the group consisting of,
the predetermined recommendation condition includes: the difference is less than or equal to a difference threshold.
9. An object recommendation apparatus comprising:
the user feature generation unit is used for generating local feature representation for any user according to the self feature representation of the associated object which has the direct association relation with the user;
the object feature generation unit is used for generating local feature representation for any object according to the self feature representation of the associated user with the direct association relation with the object;
the relation characteristic generating unit is used for generating relation characteristic representation between the user and the object according to the indirect incidence relation between the user and the object;
a feature difference calculation unit that calculates a sum of the local feature representation of the user and the relational feature representation, and calculates a difference between the sum and the local feature representation of the object;
an object recommending unit recommending the object to the user if the difference meets a predetermined recommending condition;
wherein the process of generating the local feature representation of the user and the object comprises:
generating a related object feature representation matrix according to the feature representation of each related object;
generating a correlation user characteristic representation matrix according to the self characteristic representation of each correlation user;
generating a cooperative matrix of the associated object and the associated user according to the associated object feature representation matrix and the associated user feature representation matrix;
respectively performing row pooling and column pooling on the collaborative matrix to obtain the attention vectors of the associated object and the associated user;
generating a local feature representation for the user based on the associated object feature representation matrix and the attention vector of the associated object;
generating a local feature representation for the object based on the matrix of associated user feature representations and the attention vector of the associated user.
10. The apparatus of claim 9, the generating a co-matrix of the associated objects and the associated users from the associated object feature representation matrix and the associated user feature representation matrix, comprising:
taking the correlation object feature representation matrix as input, and outputting a transformation correlation object matrix by adopting a trained first neural network model;
taking the associated user feature representation matrix as input, and outputting a transformation associated user matrix by adopting a trained second neural network model;
and calculating the product of the transformation associated object matrix and the transformation associated user matrix to obtain the cooperative matrix.
11. The apparatus of claim 9, the generating a local feature representation for the user based on the associated object feature representation matrix and the attention vector of the associated object, comprising:
taking each element value in the associated object attention vector as a weight value, and respectively carrying out weighted summation on elements which come from different associated objects and have the same dimension in the associated object feature representation matrix so as to generate local feature representation for the user;
the generating a local feature representation for the object based on the associated user feature representation matrix and the attention vector of the associated user comprises:
and taking each element value in the associated user attention vector as a weight value, and respectively carrying out weighted summation on elements which come from different associated users and have the same dimension in the associated user feature representation matrix so as to generate local feature representation for the object.
12. The apparatus of claim 9, the relationship feature generation unit to:
acquiring a plurality of indirectly reachable paths between the user and the object, wherein the indirectly reachable paths connect the user and the object through one or more nodes, connecting edges on the indirectly reachable paths represent that the connected nodes have a direct association relationship, and the types of the nodes include: user class nodes, object class nodes and non-user and non-object class nodes;
aiming at each indirect reachable path, generating the characteristic representation of the indirect reachable path according to the characteristic representation of each node on the indirect reachable path;
and fusing the characteristic representation of each indirect reachable path to obtain the relation characteristic representation between the user and the object.
13. The apparatus of claim 12, the obtaining a number of indirectly reachable paths between the user and the object, comprising:
constructing a heterogeneous network comprising user class nodes, object class nodes and non-user and non-object class nodes, wherein connecting edges in the heterogeneous network represent that the connected nodes have a direct association relationship;
randomly selecting a specified number of nodes from the heterogeneous network, and respectively starting from the selected nodes to extract paths in a random walk mode;
and acquiring an indirect reachable path between the user and the object from the extracted path.
14. The apparatus of claim 12, wherein the generating the feature representation of the indirectly reachable path according to the own feature representation of each node on the indirectly reachable path comprises:
and taking the characteristic representation of each node on the indirect reachable path as input, and outputting the characteristic representation of the indirect reachable path by adopting an LSTM model.
15. The apparatus of claim 12, said fusing the feature representation of each indirectly reachable path to obtain a relationship feature representation between the user and the object, comprising:
determining a weight value of the indirect reachable path according to the hop count of the indirect reachable path;
and according to the weight value, fusing the feature representation of each indirect reachable path in a weighted average mode to obtain the relationship feature representation between the user and the object.
16. The apparatus of claim 9, wherein the first and second electrodes are disposed on opposite sides of the substrate,
the predetermined recommendation condition includes: the difference is less than or equal to a difference threshold.
17. An object recommendation apparatus comprising:
a processor;
a memory for storing machine executable instructions;
wherein, by reading and executing machine-executable instructions stored by the memory that correspond to object recommendation logic, the processor is caused to:
aiming at any user, generating local feature representation for the user according to the feature representation of an associated object which has a direct association relation with the user;
aiming at any object, generating local feature representation for the object according to the feature representation of an associated user having a direct association relation with the object;
generating a relation characteristic representation between the user and the object according to the indirect incidence relation between the user and the object;
calculating a sum of the local feature representation of the user and the relational feature representation and calculating a difference of the sum and the local feature representation of the object;
if the difference meets a preset recommendation condition, recommending the object to the user;
wherein the process of generating the local feature representation of the user and the object comprises:
generating a related object feature representation matrix according to the feature representation of each related object;
generating a correlation user characteristic representation matrix according to the self characteristic representation of each correlation user;
generating a cooperative matrix of the associated object and the associated user according to the associated object feature representation matrix and the associated user feature representation matrix;
respectively performing row pooling and column pooling on the collaborative matrix to obtain the attention vectors of the associated object and the associated user;
generating a local feature representation for the user based on the associated object feature representation matrix and the attention vector of the associated object;
generating a local feature representation for the object based on the matrix of associated user feature representations and the attention vector of the associated user.
CN201910063963.9A 2019-01-23 2019-01-23 Object recommendation method and device Active CN110020910B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910063963.9A CN110020910B (en) 2019-01-23 2019-01-23 Object recommendation method and device

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910063963.9A CN110020910B (en) 2019-01-23 2019-01-23 Object recommendation method and device

Publications (2)

Publication Number Publication Date
CN110020910A CN110020910A (en) 2019-07-16
CN110020910B true CN110020910B (en) 2022-05-24

Family

ID=67188854

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910063963.9A Active CN110020910B (en) 2019-01-23 2019-01-23 Object recommendation method and device

Country Status (1)

Country Link
CN (1) CN110020910B (en)

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111860783B (en) * 2020-07-22 2021-07-30 中山大学 Graph node low-dimensional representation learning method and device, terminal equipment and storage medium
CN112182424B (en) * 2020-11-11 2023-01-31 重庆邮电大学 Social recommendation method based on integration of heterogeneous information and isomorphic information networks
CN114637923B (en) * 2022-05-19 2022-09-02 之江实验室 Data information recommendation method and device based on hierarchical attention-graph neural network

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108520450A (en) * 2018-03-21 2018-09-11 电子科技大学 Local low-rank matrix based on implicit feedback information approximately recommends method and system
CN108711075A (en) * 2018-05-22 2018-10-26 阿里巴巴集团控股有限公司 A kind of Products Show method and apparatus

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20180075514A1 (en) * 2016-09-15 2018-03-15 Xerox Corporation Method and system for data processing for generating adaptive recommendations of relevant items

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108520450A (en) * 2018-03-21 2018-09-11 电子科技大学 Local low-rank matrix based on implicit feedback information approximately recommends method and system
CN108711075A (en) * 2018-05-22 2018-10-26 阿里巴巴集团控股有限公司 A kind of Products Show method and apparatus

Also Published As

Publication number Publication date
CN110020910A (en) 2019-07-16

Similar Documents

Publication Publication Date Title
JP6615362B2 (en) Method and apparatus for obtaining user caricature
CN112000819B (en) Multimedia resource recommendation method and device, electronic equipment and storage medium
Zhou et al. Kernelized probabilistic matrix factorization: Exploiting graphs and side information
US20180342004A1 (en) Cumulative success-based recommendations for repeat users
CN110046301B (en) Object recommendation method and device
CN110020910B (en) Object recommendation method and device
CN107464132B (en) Similar user mining method and device and electronic equipment
WO2016015444A1 (en) Target user determination method, device and network server
CN112800342B (en) Recommendation method, system, computer device and storage medium based on heterogeneous information
CN110555050A (en) heterogeneous network node representation learning method based on meta-path
CN109903103B (en) Method and device for recommending articles
CN111340565B (en) Information recommendation method, device, equipment and storage medium
CN110689110B (en) Method and device for processing interaction event
CN111008335B (en) Information processing method, device, equipment and storage medium
CN113722546B (en) Abnormal user account acquisition method and device, electronic equipment and storage medium
CN110750238B (en) Method and device for determining product demand and electronic equipment
CN116911953A (en) Article recommendation method, apparatus, electronic device and computer readable storage medium
CN108848152A (en) A kind of method and server of object recommendation
CN115758271A (en) Data processing method, data processing device, computer equipment and storage medium
US20160335691A1 (en) Searching device, searching method, recording medium, and program
CN117112880A (en) Information recommendation and multi-target recommendation model training method and device and computer equipment
CN113297467A (en) Recommendation method, recommendation device and computer storage medium
JP6505755B2 (en) INFORMATION PROCESSING APPARATUS, INFORMATION PROCESSING METHOD, AND PROGRAM
CN113779369B (en) Matching method, matching device, electronic equipment and storage medium
EP3895032A1 (en) System and method for obtaining recommendations using scalable cross-domain collaborative filtering

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
REG Reference to a national code

Ref country code: HK

Ref legal event code: DE

Ref document number: 40012072

Country of ref document: HK

TA01 Transfer of patent application right
TA01 Transfer of patent application right

Effective date of registration: 20200925

Address after: Cayman Enterprise Centre, 27 Hospital Road, George Town, Grand Cayman Islands

Applicant after: Innovative advanced technology Co.,Ltd.

Address before: Cayman Enterprise Centre, 27 Hospital Road, George Town, Grand Cayman Islands

Applicant before: Advanced innovation technology Co.,Ltd.

Effective date of registration: 20200925

Address after: Cayman Enterprise Centre, 27 Hospital Road, George Town, Grand Cayman Islands

Applicant after: Advanced innovation technology Co.,Ltd.

Address before: A four-storey 847 mailbox in Grand Cayman Capital Building, British Cayman Islands

Applicant before: Alibaba Group Holding Ltd.

GR01 Patent grant
GR01 Patent grant