CN116108284A - Commodity recommendation method with multi-granularity attribute set cooperated with neighbor attention - Google Patents

Commodity recommendation method with multi-granularity attribute set cooperated with neighbor attention Download PDF

Info

Publication number
CN116108284A
CN116108284A CN202211436405.0A CN202211436405A CN116108284A CN 116108284 A CN116108284 A CN 116108284A CN 202211436405 A CN202211436405 A CN 202211436405A CN 116108284 A CN116108284 A CN 116108284A
Authority
CN
China
Prior art keywords
user
attribute
commodity
neighbor
collaborative
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.)
Pending
Application number
CN202211436405.0A
Other languages
Chinese (zh)
Inventor
郑建兴
景腾岳
程利涛
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.)
Shanxi University
Original Assignee
Shanxi University
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 Shanxi University filed Critical Shanxi University
Priority to CN202211436405.0A priority Critical patent/CN116108284A/en
Publication of CN116108284A publication Critical patent/CN116108284A/en
Pending legal-status Critical Current

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
    • G06F16/9536—Search customisation based on social or collaborative filtering
    • 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
    • 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
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
    • Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
    • Y02P90/30—Computing systems specially adapted for manufacturing

Landscapes

  • Engineering & Computer Science (AREA)
  • Databases & Information Systems (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Business, Economics & Management (AREA)
  • General Physics & Mathematics (AREA)
  • Accounting & Taxation (AREA)
  • General Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Finance (AREA)
  • Development Economics (AREA)
  • Economics (AREA)
  • Marketing (AREA)
  • Strategic Management (AREA)
  • General Business, Economics & Management (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention belongs to the technical field of electronic commerce commodity cold start recommendation, and discloses a commodity recommendation method of cooperative neighbor attention of a multi-granularity attribute set, which comprises the steps of firstly constructing a user-user similarity relation graph by utilizing the multi-granularity attribute set of a user, constructing a commodity-commodity similarity relation graph by utilizing the multi-granularity attribute set of a commodity, and learning cooperative neighbor embedded representation of the user and the commodity on the multi-granularity attribute set through a graph neural network; then fusing the collaborative neighbor embedded representation on the multi-granularity attribute set through an attention mechanism, and modeling a user embedded representation and a commodity embedded representation; and calculating the score of the user on the commodity by the inner product, learning the preference of the user attribute on the commodity attribute, and generating a sorted commodity recommendation list according to the score. The invention digs the personalized interests of the user from the multi-granularity attribute combination of the user and the multi-granularity attribute combination of the commodity, realizes the commodity recommendation of cold start of interactive behavior, and particularly provides great support in the aspects of new users and new commodity recommendation of electronic commerce.

Description

Commodity recommendation method with multi-granularity attribute set cooperated with neighbor attention
Technical Field
The invention belongs to the technical field of electronic commerce commodity cold start recommendation, and particularly relates to a commodity recommendation method with multi-granularity attribute set and cooperative neighbor attention.
Background
In electronic commerce commodity recommendation systems, users and commodities often have multiple attributes with a large amount of auxiliary information, and potential features of the users and the commodities can be characterized in different attribute spaces. For example, the user has characteristics of "gender", "age", "occupation", "region", and the like, and the commodity has characteristics of "color", "type", "price", "producing place", and the like. The scoring preferences of the user may be mined based on the attributes of the user, and the potential market value of the commodity may be mined based on the attributes of the commodity. Different types of attribute sets of users can construct different types of similar user groups, and different preference predictions of the users are realized. For example, users of the same "gender" are a group of people; the same user is another group for both "gender" and "age". Different types of similar commodity combinations can be constructed by the attribute sets of different types of commodities, and collaborative combination recommendation of multiple types of commodities is realized. For example, a category of goods that are "the same color" may be offered for combined sales; another type of merchandise of the same "color" and "type" at the same time may also be sold in combination. Therefore, different user-user and commodity-commodity relation diagrams can be constructed based on different types of attribute set combinations, and neighbors of users and commodities with different granularities are formed respectively. The neighbor users and the neighbor commodities based on different granularity attribute views can perform different types of feature modeling on the users and the commodities, and further score preference interests of the users on the commodities are mined from the user feature combinations of the multi-granularity attributes and the commodity feature combinations of the multi-granularity attributes.
In most recommendation systems, interactions between user items are often used to model vector representations of users and items, find similar users, and recommend similar products. Under a real recommendation scene, due to the fact that new users and new commodities are added, at the moment, the interactive historical data are less, the user and the object are difficult to fully obtain the representation, so that a collaborative recommendation system still faces the recommendation problem of cold start and data sparsity, and the prediction accuracy of scoring recommendation is low. Therefore, how to effectively utilize the rich attribute information of different types of users and commodities, realize the feature modeling of the users and the commodities, learn the grading prediction of the users on the commodities, realize accurate electronic commerce product recommendation, and especially, the recommendation performance can be improved in the face of the problems of cold start and data sparsity of the users and the commodities.
Disclosure of Invention
Aiming at the problem that the existing electronic commerce commodity recommendation system has poor recommendation precision due to the problems of cold start and data sparsity, the invention provides a commodity recommendation method with multi-granularity attribute set and cooperative neighbor attention.
In order to achieve the above purpose, the present invention adopts the following technical scheme:
the invention provides a commodity recommendation method with multi-granularity attribute set and cooperative neighbor attention, which comprises the following steps:
step S1, constructing a user-user similarity relation diagram according to a single attribute, and learning collaborative neighbor embedded representation of a user on the single attribute based on a graph neural network;
step S2, constructing a user-user similarity relation diagram according to the attribute subset, and learning collaborative neighbor embedded representation of the user on the attribute subset based on a graph neural network;
step S3, constructing a user-user similarity relation diagram according to the attribute corpus, and learning collaborative neighbor embedded representation of the user on the attribute corpus based on a graph neural network;
step S4, merging the collaborative neighbor embedded representation on the multi-granularity attribute set through an attention mechanism, and modeling a user embedded representation;
step S5, according to the steps S1-S4, the collaborative neighbor embedded representation on the multi-granularity attribute set of the commodity is fused in the same way, and the commodity embedded representation is modeled;
step S6, calculating the score of the user on the commodity through inner product calculation according to the characteristic representation of the user and the characteristic representation of the commodity;
and S7, calculating the score of the new commodity by the new user, and sorting according to the score, so as to generate a recommended commodity list and finish commodity recommendation.
Further, in the step S1, a user-user similarity relationship diagram is constructed according to a single attribute, and collaborative neighbor embedded representation of a user on the single attribute is learned based on a graph neural network, which comprises the following specific steps:
step 1.1, users with the same attribute value on a single attribute can be regarded as neighbor users under the single attribute view. Given attribute a, if user u and user u' take the same value on attribute a, it is noted as f a (u)＝f a (u'), defining the neighbor set of user u on attribute a as:
Figure BDA0003946897180000031
in the formula (1)
Figure BDA0003946897180000032
Representing the neighbor set of user u on attribute a, the number of attributes a is denoted as a.
And then, establishing a user-user attribute relation matrix U multiplied by U through the user neighbor relation on the attribute a. The user-user attribute relation matrix on the attribute a reflects coarse-grained collaborative interest preference of the user and has a certain influence on the decision of the user. In the user-user attribute relation matrix U×U, the element value is 1, which indicates that the user has the same value as the user on the attribute a.
Step 1.2, the interest of the cooperative attribute between the users can be represented by neighbor information, and the cooperative neighbor interaction information on the attribute a between the user u and the user u' is defined as follows:
Figure BDA0003946897180000033
in the formula (2)
Figure BDA0003946897180000034
Representing neighbor interaction information on attribute a between user u and user u ', u representing the embedded representation of user u, and u ' representing the embedded representation of user u '. />
Figure BDA0003946897180000035
Representing user u's neighbor set on attribute aClose and/or fill>
Figure BDA0003946897180000036
Representing the neighbor set of user u' on attribute a.
Step 1.3, considering all similar neighbors, defining collaborative user attributes of collaborative neighbors based on the graph neural network as follows:
Figure BDA0003946897180000041
in the formula (3), the amino acid sequence of the compound,
Figure BDA0003946897180000042
representing neighbor interaction information on attribute a between user u and user u'. />
Figure BDA0003946897180000043
The collaborative neighbor embedded representation of user u on a single attribute user-user similarity graph is represented.
Further, in the step S2, a user-user similarity relationship graph is constructed according to the attribute subset, and collaborative neighbor embedded representation of the user on the attribute subset is learned based on the graph neural network, which specifically comprises the following steps:
step 2.1 obtaining the neighbor set of user u on attribute A
Figure BDA0003946897180000044
And then, through the user neighbor relation on the attribute A ', a user-user attribute relation matrix U multiplied by U is established, and the element value is A ', which indicates that the value of the user on the attribute A ' is the same as that of the user. Thereby constructing a user-user similarity relation diagram of the attribute subset.
Step 2.2, defining cooperative neighbor interaction information on attribute a 'between user u and user u' as:
Figure BDA0003946897180000045
in (4)
Figure BDA0003946897180000046
Representing neighbor interaction information on attribute A 'between user u and user u', u representing the embedded representation of user u, and u 'representing the embedded representation of user u'. />
Figure BDA0003946897180000047
Representing the neighbor set of user u on attribute A', ∈>
Figure BDA0003946897180000048
Representing the neighbor set of user u 'on attribute a'.
Step 2.3, considering all similar neighbors, defining collaborative user attributes of collaborative neighbors based on the graph neural network as follows:
Figure BDA0003946897180000049
in the formula (5), the amino acid sequence of the compound,
Figure BDA00039468971800000410
representing neighbor interaction information on attribute a 'between user u and user u'. />
Figure BDA00039468971800000411
The collaborative neighbor embedded representation of user u on the attribute subset user-user similarity relationship graph is represented.
Further, in the step S3, a user-user similarity relation diagram is constructed according to the attribute corpus, and collaborative neighbor embedded representation of the user on the attribute corpus is learned based on a graph neural network, which comprises the following specific steps:
step 3.1, acquiring neighbor set of user u on attribute A
Figure BDA0003946897180000051
And then, through the user neighbor relation on the attribute A, a user-user attribute relation matrix U multiplied by U is established, wherein the element value is A, and the element value is the same as the value of the user on the attribute A. Thereby constructing attribute corpus user-user similarity relationshipIs a drawing.
Step 3.2, defining cooperative neighbor interaction information on the attribute A between the user u and the user u' as follows:
Figure BDA0003946897180000052
in (6)
Figure BDA0003946897180000053
Representing neighbor interaction information on attribute a between user u and user u ', u representing the embedded representation of user u, and u ' representing the embedded representation of user u '. />
Figure BDA0003946897180000054
Representing the neighbor set of user u on attribute A, < >>
Figure BDA0003946897180000055
Representing the neighbor set of user u' on attribute a.
Step 3.3, considering all similar neighbors, defining collaborative user attributes of collaborative neighbors based on the graph neural network as follows:
Figure BDA0003946897180000056
in the formula (7), the amino acid sequence of the compound,
Figure BDA0003946897180000057
representing neighbor interaction information on attribute a between user u and user u'. />
Figure BDA0003946897180000058
Representing a collaborative neighbor embedded representation of user u on a property corpus user-user similarity graph.
Further, in the step S4, collaborative neighbor embedded representations on the multi-granularity attribute set are fused through an attention mechanism, and the user embedded representation is modeled, which specifically includes the steps of:
step 4.1, user-basedWe can get multiple collaborative-neighbor embedded representations of the user's attribute sets (a, a', …, a)
Figure BDA0003946897180000059
Step 4.2, on the attribute a view, computing a collaborative neighbor embedding representation
Figure BDA00039468971800000512
The output feature score of (2) is:
Figure BDA00039468971800000510
in formula (8), W h H is a weight parameter of the neural network;
step 4.3, defining a collaborative neighbor embedding representation
Figure BDA00039468971800000511
The attention weights of (2) are:
Figure BDA0003946897180000061
in the formula (9), the amino acid sequence of the compound,
Figure BDA0003946897180000069
embedding normalized attention weights for collaborative neighbors of users on the attribute a view, exp being an exponential function;
and 4.4, fusing collaborative neighbor embedments of the user on all attribute views by using an attention mechanism, wherein the definition of the user embedments is expressed as follows:
Figure BDA0003946897180000062
in the formula (10), the amino acid sequence of the compound,
Figure BDA0003946897180000063
for collaborative neighbors on different attribute viewsThe extent to which the embedding contributes to the user's embedded representation, i, is a subset of the user's attributes.
Further, in the step S5, according to the steps S1 to S4, the collaborative neighbor embedded representation on the multi-granularity attribute set of the commodity is fused, and the commodity embedded representation is modeled, specifically comprising the following steps:
step 5.1, obtaining the commodity expression of the commodity v on the single attribute commodity-commodity similarity relation view by the same method as the step 1
Figure BDA0003946897180000064
b represents a single attribute.
Step 5.2, obtaining the commodity expression of the commodity v on the attribute subset commodity-commodity similarity relation view by the same method as the step 2
Figure BDA0003946897180000065
B' represents an attribute in the subset of attributes.
Step 5.3, obtaining the commodity expression of the commodity v on the attribute corpus commodity-commodity similarity relation view by the same method as the step 3
Figure BDA0003946897180000066
B represents an attribute in the attribute corpus.
Step 5.4, obtaining the final embedded representation e of the commodity by the same method as the step S4 v 。
Further, in the step S6, the score of the user to the commodity is calculated according to the characteristic representation of the user and the characteristic representation of the commodity by the inner product, which specifically comprises the following steps:
we do an inner product to estimate the user's score for the good. The specific scoring formula is as follows:
Figure BDA0003946897180000067
in the formula (11), the amino acid sequence of the compound,
Figure BDA0003946897180000068
representing the user u versus the commodity vAnd final predictive scoring. b g Representing global bias entries, b u Representing bias terms of a user, b v The bias term representing the item.
Further, in the step S7, the score of the new user on the new commodity is calculated, and the commodity recommendation list is generated according to the ranking of the scores, so as to complete commodity recommendation, which comprises the following specific steps:
for new user u n The new user u is acquired in the same way as in steps S1-S3 n Is embedded in the representation of (a)
Figure BDA0003946897180000071
For new commodity v m The new commodity v is obtained by the same method as in step S4 m Is embedded in the representation +.>
Figure BDA0003946897180000072
Obtaining a new user u according to equation (11) n For new commodity v m And the final predictive score y of (2). And sorting the commodities in the candidate set according to the score, generating a recommended commodity list, and completing commodity recommendation.
The invention also provides a computer readable storage medium, wherein the medium is stored with a computer program, and the computer program is executed to realize the commodity recommendation method of the multi-granularity attribute set collaborative neighbor attention.
The invention also provides an electronic device, which comprises a memory, a processor and a computer program stored on the memory and capable of running on the processor, wherein the processor realizes the commodity recommendation method of the multi-granularity attribute set collaborative neighbor attention when executing the computer program.
Compared with the prior art, the invention has the following advantages:
the method provided by the invention is distinguished from the existing method by the following remarkable characteristics: potential preferences of users and commodities are mined through different multi-granularity attribute sets, semantic embedding of the users and the commodities is constructed, and a scoring prediction model based on semantic representation of the multi-granularity attribute sets of the users and the commodities is established. The method and the device effectively improve the recommendation precision, simultaneously relieve the problem of low recommendation performance in a sparse situation, and realize the commodity recommendation of cold start of interaction behavior.
Drawings
FIG. 1 is a schematic diagram of an overall model architecture of the present invention;
FIG. 2 is a comparison of the present invention with other neural network methods on ML-100kr datasets under different sparsity conditions.
Detailed Description
The commodity recommendation method of the multi-granularity attribute set collaborative neighbor attention is implemented through a computer program. The following describes the specific embodiments of the technical scheme according to the present invention in detail according to the flow.
As shown in fig. 1, the commodity recommendation method of the multi-granularity attribute set collaborative neighbor attention of the present invention includes the following steps:
step S1, constructing a user-user similarity relation diagram according to a single attribute, and learning collaborative neighbor embedded representation of a user on the single attribute based on a graph neural network, wherein the specific steps are as follows:
step 1.1, users with the same attribute value on a single attribute can be regarded as neighbor users under a single attribute view. Given attribute a, if user u and user u' take the same value on attribute a, it is noted as f a (u)＝f a (u'), defining the neighbor set of user u on attribute a as:
Figure BDA0003946897180000081
in the formula (1)
Figure BDA0003946897180000082
Representing the neighbor set of user u on attribute a, the number of attributes a is denoted as a.
And then, establishing a user-user attribute relation matrix U multiplied by U through the user neighbor relation on the attribute a. The user-user attribute relation matrix on the attribute a reflects coarse-grained collaborative interest preference of the user and has a certain influence on the decision of the user. In the user-user attribute relation matrix U×U, the element value is 1, which indicates that the user has the same value as the user on the attribute a.
Step 1.2, the interest of the cooperative attribute between the users can be represented by neighbor information, and the cooperative neighbor interaction information on the attribute a between the user u and the user u' is defined as follows:
Figure BDA0003946897180000083
in the formula (2)
Figure BDA0003946897180000084
Representing neighbor interaction information on attribute a between user u and user u ', u representing the embedded representation of user u, and u ' representing the embedded representation of user u '. />
Figure BDA0003946897180000085
Representing the neighbor set of user u on attribute a, +.>
Figure BDA0003946897180000086
Representing the neighbor set of user u' on attribute a.
Step 1.3, considering all similar neighbors, defining collaborative user attributes of collaborative neighbors based on the graph neural network as follows:
Figure BDA0003946897180000091
in the formula (3), the amino acid sequence of the compound,
Figure BDA0003946897180000092
representing neighbor interaction information on attribute a between user u and user u'. />
Figure BDA0003946897180000093
The collaborative neighbor embedded representation of user u on a single attribute user-user similarity graph is represented.
Step S2, constructing a user-user similarity relation diagram according to the attribute subset, and learning collaborative neighbor embedded representation of the user on the attribute subset based on a graph neural network, wherein the specific steps are as follows:
step 2.1, acquiring neighbor set of user u on attribute A
Figure BDA0003946897180000094
And then, through the user neighbor relation on the attribute A ', a user-user attribute relation matrix U multiplied by U is established, and the element value is A ', which indicates that the value of the user on the attribute A ' is the same as that of the user. Thereby constructing a user-user similarity relation diagram of the attribute subset.
Step 2.2, defining cooperative neighbor interaction information on attribute a 'between user u and user u' as:
Figure BDA0003946897180000095
in (4)
Figure BDA0003946897180000096
Representing neighbor interaction information on attribute A 'between user u and user u', u representing the embedded representation of user u, and u 'representing the embedded representation of user u'. />
Figure BDA0003946897180000097
Representing the neighbor set of user u on attribute A', ∈>
Figure BDA0003946897180000098
Representing the neighbor set of user u 'on attribute a'.
Step 2.3, considering all similar neighbors, defining collaborative user attributes of collaborative neighbors based on the graph neural network as follows:
Figure BDA0003946897180000099
in the formula (5), the amino acid sequence of the compound,
Figure BDA00039468971800000910
representing neighbor interaction information on attribute a 'between user u and user u'. />
Figure BDA00039468971800000911
The collaborative neighbor embedded representation of user u on the attribute subset user-user similarity relationship graph is represented.
Step S3, constructing a user-user similarity relation diagram according to the attribute corpus, and learning collaborative neighbor embedded representation of a user on the attribute corpus based on a graph neural network, wherein the specific steps are as follows:
step 3.1, acquiring neighbor set of user u on attribute A
Figure BDA0003946897180000101
And then, through the user neighbor relation on the attribute A, a user-user attribute relation matrix U multiplied by U is established, wherein the element value is A, and the element value is the same as the value of the user on the attribute A. Thus constructing the attribute corpus user-user similarity relation graph.
Step 3.2, defining cooperative neighbor interaction information on the attribute A between the user u and the user u' as follows:
Figure BDA0003946897180000102
in (6)
Figure BDA0003946897180000103
Representing neighbor interaction information on attribute a between user u and user u ', u representing the embedded representation of user u, and u ' representing the embedded representation of user u '. />
Figure BDA0003946897180000104
Representing the neighbor set of user u on attribute A, < >>
Figure BDA0003946897180000105
Representing the neighbor set of user u' on attribute a.
Step 3.3, considering all similar neighbors, defining collaborative user attributes of collaborative neighbors based on the graph neural network as follows:
Figure BDA0003946897180000106
/>
in the formula (7), the amino acid sequence of the compound,
Figure BDA0003946897180000107
representing neighbor interaction information on attribute a between user u and user u'. />
Figure BDA0003946897180000108
Representing a collaborative neighbor embedded representation of user u on a property corpus user-user similarity graph.
Step S4, fusing collaborative neighbor embedded representations on the multi-granularity attribute set through an attention mechanism, and modeling user embedded representations, wherein the specific steps are as follows:
step 4.1, based on the user's attribute set (a, A', …, A), we can get multiple collaborative-neighbor embedded representations of the user
Figure BDA0003946897180000109
Step 4.2, on the attribute a view, computing a collaborative neighbor embedding representation
Figure BDA00039468971800001010
The output feature score of (2) is:
Figure BDA00039468971800001011
in formula (8), W h H is a weight parameter of the neural network;
step 4.3, defining a collaborative neighbor embedding representation
Figure BDA00039468971800001012
The attention weights of (2) are:
Figure BDA0003946897180000111
in the formula (9), the amino acid sequence of the compound,
Figure BDA0003946897180000112
embedding normalized attention weights for collaborative neighbors of users on the attribute a view, exp being an exponential function;
and 4.4, fusing collaborative neighbor embedments of the user on all attribute views by using an attention mechanism, wherein the definition of the user embedments is expressed as follows:
Figure BDA0003946897180000113
in the formula (10), the amino acid sequence of the compound,
Figure BDA0003946897180000114
the extent to which collaborative neighbor embedding contributes to the user embedding representation for different attribute views, i is the attribute subset of the user.
Step S5, according to the steps S1-S4, the collaborative neighbor embedded representation on the multi-granularity attribute set of the commodity is fused, and the commodity embedded representation is modeled, wherein the specific steps are as follows:
step 5.1, obtaining the commodity expression of the commodity v on the single attribute commodity-commodity similarity relation view by the same method as the step 1
Figure BDA0003946897180000115
b represents a single attribute.
Step 5.2, obtaining the commodity expression of the commodity v on the attribute subset commodity-commodity similarity relation view by the same method as the step 2
Figure BDA0003946897180000116
B' represents an attribute in the subset of attributes.
Step 5.3, obtaining the commodity expression of the commodity v on the attribute corpus commodity-commodity similarity relation view by the same method as the step 3
Figure BDA0003946897180000117
B represents an attribute in the attribute corpus.
Step 5.4, obtaining the final embedded representation e of the commodity by the same method as the step S4 v 。
Step S6, calculating the score of the commodity by the user according to the characteristic representation of the user and the characteristic representation of the commodity, and constructing a score prediction model, wherein the specific steps are as follows:
we do an inner product to estimate the user's score for the good. The specific scoring formula is as follows:
Figure BDA0003946897180000118
in the formula (11), the amino acid sequence of the compound,
Figure BDA0003946897180000119
representing the final predictive score of user u for commodity v. b g Representing global bias entries, b u Representing bias terms of a user, b v The bias term representing the item.
Step S7, calculating the score of the new commodity by the new user, and sorting according to the score, so as to generate a recommended commodity list, wherein the specific steps are as follows:
for new user u n The new user u is acquired in the same way as in steps S1-S3 n Is embedded in the representation of (a)
Figure BDA0003946897180000121
For new commodity v m The new commodity v is obtained by the same method as in step S4 m Is embedded in the representation +.>
Figure BDA0003946897180000122
Obtaining a new user u according to equation (11) n For new commodity v m And the final predictive score y of (2). And sorting the commodities in the candidate set according to the score, generating a recommended commodity list, and completing commodity recommendation.
To verify the effectiveness of the method, we performed experiments on the ML-100kr dataset (https:// groups tens. Org/data/movieens /), the dataset information being shown in Table 1:
table 1 dataset case
Figure BDA0003946897180000123
Data sets are processed according to training set, verification set and test set 8:1:1 division experiments were performed using RMSE as an evaluation index. In order to verify the effectiveness and the advancement of the technical scheme provided by the invention, several existing scoring recommendation prediction model methods are selected for comparison: GCN, NGCF, GAT, lightGCN. The experimental results are shown in table 2:
table 2 experimental results
Figure BDA0003946897180000124
As can be seen from the results in Table 2, the technical scheme of the invention can obtain the detection results with accuracy and reliability superior to those of the existing method when the scoring prediction of the commodity by the user is carried out.
Meanwhile, the invention sets sparsity of different proportions on the data by randomly shielding the grading labels, and observes the performance of the invention and the corresponding graphic neural network on the RMSE index on the processed data set. The experimental results are shown in FIG. 2. From fig. 2, it can be seen that as the data sparsity increases, the RMSE index of several methods is significantly increased, which indicates that the accuracy of the user's scoring prediction for the merchandise is decreasing, indicating that sparsity can affect the quality of the embedded representation of the user and item. Meanwhile, the method provided by the invention can be found that the RMSE index is lower than that of the corresponding graph neural network method under the condition of data sparsity of different proportions, and can show better scoring prediction performance.
Example 2
The present embodiment provides a computer-readable storage medium having stored thereon a computer program that, when executed, implements the multi-granularity attribute set collaborative neighbor attention commodity recommendation method of embodiment 1 described above.
Example 3
The present embodiment provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor, where the processor implements the multi-granularity attribute set collaborative neighbor attention commodity recommendation method of embodiment 1.
The foregoing examples are illustrative of the present invention and are not intended to be limiting, and it will be apparent to those skilled in the art that various modifications and improvements can be made without departing from the spirit of the invention.
Furthermore, it should be understood that although the present disclosure describes embodiments, not every embodiment is provided with a separate embodiment, and that this description is provided for clarity only, and that the disclosure is not limited to the embodiments described in detail below, and that the embodiments described in the examples may be combined as appropriate to form other embodiments that will be apparent to those skilled in the art.

Claims (10)

1. The commodity recommendation method of the multi-granularity attribute set collaborative neighbor attention is characterized by comprising the following steps of: the method comprises the following steps:
step S1, constructing a user-user similarity relation diagram according to a single attribute, and learning collaborative neighbor embedded representation of a user on the single attribute based on a graph neural network;
step S2, constructing a user-user similarity relation diagram according to the attribute subset, and learning collaborative neighbor embedded representation of the user on the attribute subset based on a graph neural network;
step S3, constructing a user-user similarity relation diagram according to the attribute corpus, and learning collaborative neighbor embedded representation of the user on the attribute corpus based on a graph neural network;
step S4, merging the collaborative neighbor embedded representation on the multi-granularity attribute set through an attention mechanism, and modeling a user embedded representation;
step S5, according to the steps S1-S4, the collaborative neighbor embedded representation on the multi-granularity attribute set of the commodity is fused in the same way, and the commodity embedded representation is modeled;
step S6, calculating the score of the user on the commodity according to the characteristic representation of the user and the characteristic representation of the commodity through the inner product;
and S7, calculating the score of the new commodity by the new user, and sorting according to the score, so as to generate a recommended commodity list and finish commodity recommendation.
2. The commodity recommendation method for multi-granularity attribute set collaborative neighbor attention according to claim 1, wherein the commodity recommendation method comprises the following steps: in the step S1, a user-user similarity relation diagram is constructed according to a single attribute, and the specific steps of learning collaborative neighbor embedded representation of a user on the single attribute based on a graph neural network are as follows:
step 1.1, the users with the same attribute value on the single attribute can be regarded as neighbor users in the single attribute view, given attribute a, if the attribute a is the same as the attribute u', the attribute a is marked as f a (u)＝f a (u'), defining the neighbor set of user u on attribute a as:
Figure FDA0003946897170000011
in the formula (1)
Figure FDA0003946897170000021
Representing a neighbor set of the user u on the attribute a, wherein the number of the attribute a is represented by the number of the attribute a;
further, establishing a user-user attribute relation matrix U X U through the user neighbor relation on the attribute a; the user-user attribute relation matrix on the attribute a reflects coarse-grained collaborative interest preference of the user and has a certain influence on the decision of the user; in the user-user attribute relation matrix U multiplied by U, the element value is 1, which indicates that the value of the user on the attribute a is the same as that of the user;
step 1.2, the interest of the cooperative attribute between the users can be represented by neighbor information, and the cooperative neighbor interaction information on the attribute a between the user u and the user u' is defined as follows:
Figure FDA0003946897170000022
in the formula (2)
Figure FDA0003946897170000023
Representing neighbor interaction information on attribute a between user u and user u ', u representing an embedded representation of user u, u ' representing an embedded representation of user u '; />
Figure FDA0003946897170000024
Representing the neighbor set of user u on attribute a, +.>
Figure FDA0003946897170000025
Representing a neighbor set of user u' on attribute a;
step 1.3, considering all similar neighbors, defining collaborative user attributes of collaborative neighbors based on the graph neural network as follows:
Figure FDA0003946897170000026
in the formula (3), the amino acid sequence of the compound,
Figure FDA0003946897170000027
representing neighbor interaction information on attribute a between user u and user u'; />
Figure FDA0003946897170000028
The collaborative neighbor embedded representation of user u on a single attribute user-user similarity graph is represented.
3. The commodity recommendation method for multi-granularity attribute set collaborative neighbor attention according to claim 1, wherein the commodity recommendation method comprises the following steps: in the step S2, a user-user similarity relation diagram is constructed according to the attribute subset, and the specific steps of learning collaborative neighbor embedded representation of the user on the attribute subset based on the graph neural network are as follows:
step 2.1, acquiring neighbor set of user u on attribute A
Figure FDA0003946897170000029
Further, through the user neighbor relation on the attribute A ', a user-user attribute relation matrix U multiplied by U is established, the element value is A ', and the value of the user on the attribute A ' is the same, so that an attribute subset user-user similarity relation diagram is constructed;
step 2.2, defining cooperative neighbor interaction information on attribute a 'between user u and user u' as:
Figure FDA0003946897170000031
in (4)
Figure FDA0003946897170000032
Representing neighbor interaction information on an attribute A 'between a user u and a user u', wherein u represents an embedded representation of the user u, and u 'represents an embedded representation of the user u'; />
Figure FDA0003946897170000033
Representing a neighbor set of user u on attribute A'; />
Figure FDA0003946897170000034
Representing a neighbor set of user u 'on attribute A';
step 2.3, considering all similar neighbors, defining collaborative user attributes of collaborative neighbors based on the graph neural network as follows:
Figure FDA0003946897170000035
in the formula (5), the amino acid sequence of the compound,
Figure FDA0003946897170000036
representing neighbor interaction information on an attribute A 'between a user u and a user u'; />
Figure FDA0003946897170000037
The collaborative neighbor embedded representation of user u on the attribute subset user-user similarity relationship graph is represented.
4. The commodity recommendation method for multi-granularity attribute set collaborative neighbor attention according to claim 1, wherein the commodity recommendation method comprises the following steps: in the step S3, a user-user similarity relation diagram is constructed according to the attribute corpus, and the specific steps of learning collaborative neighbor embedded representation of the user on the attribute corpus based on a graph neural network are as follows:
step 3.1, acquiring neighbor set of user u on attribute A
Figure FDA0003946897170000038
Further, through the user neighbor relation on the attribute A, a user-user attribute relation matrix U X U is established, wherein the element value is A, and the value of the user on the attribute A is the same as that of the user; thus constructing the attribute corpus user-user similarity relation graph.
Step 3.2, defining cooperative neighbor interaction information on the attribute A between the user u and the user u' as follows:
Figure FDA0003946897170000039
in (6)
Figure FDA00039468971700000310
Representing neighbor interaction information on an attribute A between a user u and a user u ', wherein u represents an embedded representation of the user u, and u ' represents an embedded representation of the user u '; />
Figure FDA00039468971700000311
Representing the neighbor set of user u on attribute A, < >>
Figure FDA00039468971700000312
Representing the neighbor set of user u' on attribute A;
step 3.3, considering all similar neighbors, defining collaborative user attributes of collaborative neighbors based on the graph neural network as follows:
Figure FDA0003946897170000041
in the formula (7), the amino acid sequence of the compound,
Figure FDA0003946897170000042
representing neighbor interaction information on attribute A between user u and user u'; />
Figure FDA0003946897170000043
Representing a collaborative neighbor embedded representation of user u on a property corpus user-user similarity graph.
5. The commodity recommendation method for multi-granularity attribute set collaborative neighbor attention according to claim 1, wherein the commodity recommendation method comprises the following steps: in the step S4, collaborative neighbor embedded representations on the multi-granularity attribute set are fused through an attention mechanism, and the specific steps of modeling the user embedded representation are as follows:
step 4.1, based on the user's attribute set (a, A', …, A), we can get multiple collaborative-neighbor embedded representations of the user
Figure FDA0003946897170000044
Step 4.2, on the attribute a view, computing a collaborative neighbor embedding representation
Figure FDA0003946897170000045
The output feature score of (2) is:
Figure FDA0003946897170000046
in formula (8), W h H is a weight parameter of the neural network;
step 4.3, defining a collaborative neighbor embedding representation
Figure FDA0003946897170000047
The attention weights of (2) are:
Figure FDA0003946897170000048
in the formula (9), the amino acid sequence of the compound,
Figure FDA0003946897170000049
embedding normalized attention weights for collaborative neighbors of users on the attribute a view, exp being an exponential function;
and 4.4, fusing collaborative neighbor embedments of the user on all attribute views by using an attention mechanism, wherein the definition of the user embedments is expressed as follows:
Figure FDA00039468971700000410
in the formula (10), the amino acid sequence of the compound,
Figure FDA00039468971700000411
the extent to which collaborative neighbor embedding contributes to the user embedding representation for different attribute views, i is the attribute subset of the user.
6. The commodity recommendation method for multi-granularity attribute set collaborative neighbor attention according to claim 1, wherein the commodity recommendation method comprises the following steps: in step S5, according to steps S1-S4, the collaborative neighbor embedded representation on the multi-granularity attribute set of the commodity is fused in the same way, and the modeling of the commodity embedded representation comprises the following specific steps:
step 5.1, obtaining the commodity expression of the commodity v on the single attribute commodity-commodity similarity relation view by the same method as the step 1
Figure FDA0003946897170000051
b represents a single attribute;
step 5.2, obtaining the commodity expression of the commodity v on the attribute subset commodity-commodity similarity relation view by the same method as the step 2
Figure FDA0003946897170000052
B' represents an attribute in the subset of attributes;
step 5.3, obtaining the commodity expression of the commodity v on the attribute corpus commodity-commodity similarity relation view by the same method as the step 3
Figure FDA0003946897170000053
B represents the attributes in the attribute complete set;
step 5.4, obtaining the final embedded representation e of the commodity by the same method as the step S4 v 。
7. The commodity recommendation method for multi-granularity attribute set collaborative neighbor attention according to claim 1, wherein the commodity recommendation method comprises the following steps: in step S6, according to the feature representation of the user and the feature representation of the commodity, the specific steps of calculating the score of the user on the commodity through the inner product are as follows:
performing inner product to estimate the score of the commodity by the user; the specific scoring formula is as follows:
Figure FDA0003946897170000054
/>
in the formula (11), the amino acid sequence of the compound,
Figure FDA0003946897170000055
representing the final predictive score of user u on commodity v, b g Representing global bias entries, b u Representing bias terms of a user, b v The bias term representing the item.
8. The commodity recommendation method for multi-granularity attribute set collaborative neighbor attention according to claim 1, wherein the commodity recommendation method comprises the following steps: in step S7, calculating the score of the new commodity by the new user, sorting according to the score, generating a recommended commodity list, and completing the specific steps of commodity recommendation:
for new user u n The new user u is acquired in the same way as in steps S1-S3 n Is embedded in the representation of (a)
Figure FDA0003946897170000056
For new commodity v m The new commodity v is obtained by the same method as in step S4 m Is embedded in the representation +.>
Figure FDA0003946897170000057
According to the scoring formula->
Figure FDA0003946897170000058
Acquisition of New user u n For new commodity v m The final predictive score y of (2); and sorting the commodities in the candidate set according to the score, generating a recommended commodity list, and completing commodity recommendation.
9. A computer-readable storage medium, characterized by: the medium has stored thereon a computer program which, when executed, implements a multi-granularity attribute set collaborative neighbor attention commodity recommendation method as claimed in any one of claims 1-8.
10. An electronic device, characterized in that: a computer program comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the multi-granularity attribute set collaborative neighbor attention commodity recommendation method of any one of claims 1-8 when the computer program is executed.
CN202211436405.0A 2022-11-16 2022-11-16 Commodity recommendation method with multi-granularity attribute set cooperated with neighbor attention Pending CN116108284A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202211436405.0A CN116108284A (en) 2022-11-16 2022-11-16 Commodity recommendation method with multi-granularity attribute set cooperated with neighbor attention

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202211436405.0A CN116108284A (en) 2022-11-16 2022-11-16 Commodity recommendation method with multi-granularity attribute set cooperated with neighbor attention

Publications (1)

Publication Number Publication Date
CN116108284A true CN116108284A (en) 2023-05-12

Family

ID=86262718

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202211436405.0A Pending CN116108284A (en) 2022-11-16 2022-11-16 Commodity recommendation method with multi-granularity attribute set cooperated with neighbor attention

Country Status (1)

Country Link
CN (1) CN116108284A (en)

Similar Documents

Publication Publication Date Title
US20210271975A1 (en) User tag generation method and apparatus, storage medium, and computer device
CN111191092B (en) Label determining method and label determining model training method
Ba et al. Clustering collaborative filtering recommendation system based on SVD algorithm
CN111626832B (en) Product recommendation method and device and computer equipment
Liang et al. A probabilistic rating auto-encoder for personalized recommender systems
US20210366006A1 (en) Ranking of business object
Eliyas et al. Recommendation systems: Content-based filtering vs collaborative filtering
US11599927B1 (en) Artificial intelligence system using deep neural networks for pairwise character-level text analysis and recommendations
CN113222711B (en) Commodity information recommendation method, system and storage medium
CN111310038B (en) Information recommendation method and device, electronic equipment and computer-readable storage medium
CN114386513A (en) Interactive grading prediction method and system integrating comment and grading
CN108920647A (en) Low-rank matrix based on spectral clustering fills TOP-N recommended method
CN105868422B (en) A kind of collaborative filtering recommending method based on elastic dimensional feature vector Optimizing Extraction
CN111915414B (en) Method and device for displaying target object sequence to target user
CN114238758A (en) User portrait prediction method based on multi-source cross-border data fusion
Ifada et al. Do-rank: DCG optimization for learning-to-rank in tag-based item recommendation systems
CN116108284A (en) Commodity recommendation method with multi-granularity attribute set cooperated with neighbor attention
Qiu A predictive model for customer purchase behavior in e-commerce context
CN113254775A (en) Credit card product recommendation method based on client browsing behavior sequence
CN112765465A (en) User map-based recommendation method
Haddad et al. Towards a new model for context-aware recommendation
Aslanyan et al. Lda-lfm: A joint exploitation of review text and ratings in recommender systems
WO2023079703A1 (en) Processing execution system, processing execution method, and program
Beukman Improving collaborative filtering with fuzzy clustering
Zayet et al. A new weighting algorithm for 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