CN113868537A - Recommendation method based on multi-behavior session graph fusion - Google Patents

Recommendation method based on multi-behavior session graph fusion Download PDF

Info

Publication number
CN113868537A
CN113868537A CN202111212853.8A CN202111212853A CN113868537A CN 113868537 A CN113868537 A CN 113868537A CN 202111212853 A CN202111212853 A CN 202111212853A CN 113868537 A CN113868537 A CN 113868537A
Authority
CN
China
Prior art keywords
item
behavior
graph
session
target user
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202111212853.8A
Other languages
Chinese (zh)
Other versions
CN113868537B (en
Inventor
王英龙
张洪彪
舒明雷
陈达
刘丽
孔祥龙
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.)
Qilu University of Technology
Shandong Computer Science Center National Super Computing Center in Jinan
Shandong Institute of Artificial Intelligence
Original Assignee
Qilu University of Technology
Shandong Computer Science Center National Super Computing Center in Jinan
Shandong Institute of Artificial Intelligence
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 Qilu University of Technology, Shandong Computer Science Center National Super Computing Center in Jinan, Shandong Institute of Artificial Intelligence filed Critical Qilu University of Technology
Priority to CN202111212853.8A priority Critical patent/CN113868537B/en
Publication of CN113868537A publication Critical patent/CN113868537A/en
Application granted granted Critical
Publication of CN113868537B publication Critical patent/CN113868537B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • 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
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0241—Advertisements
    • G06Q30/0251—Targeted advertisements
    • G06Q30/0255—Targeted advertisements based on user history
    • 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
    • G06F16/9538—Presentation of query results
    • 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
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0241—Advertisements
    • G06Q30/0277—Online advertisement
    • 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)
  • Business, Economics & Management (AREA)
  • Theoretical Computer Science (AREA)
  • Databases & Information Systems (AREA)
  • Finance (AREA)
  • Accounting & Taxation (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Strategic Management (AREA)
  • Development Economics (AREA)
  • Marketing (AREA)
  • General Business, Economics & Management (AREA)
  • Economics (AREA)
  • General Engineering & Computer Science (AREA)
  • Game Theory and Decision Science (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Data Mining & Analysis (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

A recommendation method based on multi-behavior session graph fusion is used for constructing a multi-behavior undirected session graph by using combined multi-behavior sequence data of a target user and similar users. On the basis, neighbor information is aggregated according to different weights to obtain multi-behavior embedding of the project, multi-behavior embedding of the project is connected in series, and a user interest representation is obtained by combining an attention mechanism. And finally, carrying out inner product by using item embedding and user interest representation to obtain a normalized score to determine whether to recommend the item. Compared with other conversation type recommendation methods, the method has the advantages that the project embedding containing more behavior intentions can be obtained from modeling user multi-behavior sequence data; secondly, the sequence is constructed into a weighted undirected graph, so that the one-way constraint during neighbor information aggregation is removed, and the model can learn the two-way relation among projects; and thirdly, the similar users are used for supplementing the target user data, and the model can learn novel items which do not appear in the target user historical data, so that the diversity and the accuracy of the recommendation result are improved.

Description

Recommendation method based on multi-behavior session graph fusion
Technical Field
The invention relates to the technical field of session graph recommendation, in particular to a recommendation method based on multi-behavior session graph fusion.
Background
Efficient capture of user interests is the core of accurate recommendations. In life, the interest points of the users are dynamically changed, and the interest points are not only influenced by long-term interest, but also have inseparable connection with recent interactive behaviors. The traditional recommendation system carries out static recommendation according to user interaction historical behaviors, and cannot capture dynamic preference of a user. Conversational recommendation systems aim to describe the dynamic interests of users, which predict items that may interact in the future based on user click sequences, but they still face the following problems:
1. only a single type of user behavior data is used for training, and the future behavior activity of the user is ignored and is jointly determined by various historical interactive behaviors;
2. only the behavior characteristics of a single user are concerned, and the internal relation of behavior patterns among similar users is ignored;
3. in the same behavior session, the user interaction items are highly homogeneous, the model is difficult to learn novel items, and the diversity of recommendation results is reduced.
Disclosure of Invention
In order to overcome the defects of the technology, the invention provides a recommendation method based on multi-behavior session graph fusion, which improves the accuracy and diversity of recommendation.
The technical scheme adopted by the invention for overcoming the technical problems is as follows:
a recommendation method based on multi-behavior session graph fusion comprises the following steps:
a) obtaining a target user utClicking, collecting and purchasing historical conversation data and item sets in a certain shopping platform through a browser or a client
Figure BDA0003309413780000011
b) By the formula
Figure BDA0003309413780000012
Computing similarity between users
Figure BDA0003309413780000013
In the formula
Figure BDA0003309413780000014
For using a set of items interacted with by a target user
Figure BDA0003309413780000015
The embedded vector of the initialization is then set to zero,
Figure BDA0003309413780000021
for other users' embedded vectors, all other users are based on similarity
Figure BDA0003309413780000022
The values of the users are sorted in descending order, and the first N users are taken as the neighbor users u according to the sorting resultsObtaining neighbor user usClicking, collecting and purchasing historical conversation data and item sets in a certain shopping platform through a browser or a client
Figure BDA0003309413780000023
c) By the formula
Figure BDA0003309413780000024
Calculating to obtain a joint session project set of a target user and a neighbor user
Figure BDA0003309413780000025
To federate a set of session items
Figure BDA0003309413780000026
The item in (1) is a node, and a right undirected click conversation graph is constructed
Figure BDA0003309413780000027
Building collection conversation chart with right and no direction
Figure BDA0003309413780000028
And constructing a purchase session graph with rights and directions
Figure BDA0003309413780000029
d) Click on conversation graph
Figure BDA00033094137800000210
Collection session graph
Figure BDA00033094137800000211
And purchase session directed graph
Figure BDA00033094137800000212
Inputting the three behavior embedded vectors into a LightGCN network to respectively obtain three behavior embedded vectors e of item i aggregation high-order neighbor characteristicsi,click、ei,collectAnd ei,buyBy the formula ei＝ei,click||ei,collect||ei,buyCalculating to obtain a final project embedding vector eiIn the formula, | | is splicing operation;
e) by the formula
Figure BDA00033094137800000213
Calculating to obtain a recent interest preference vector g of the target user, wherein
Figure BDA00033094137800000214
For the embedded vector of the d-th item i in the current conversation sequence S of the target user,
Figure BDA00033094137800000215
LeakyReLU (-) is a non-linear activation,
Figure BDA00033094137800000216
is a vector of weights, W is a shared parameter matrix,
Figure BDA00033094137800000217
is the embedding vector of the kth item i in the current conversation sequence S of the target user, k belongs to { 1., D }, D is the length of the current conversation sequence S of the target user, e|D|Is a target user utEmbedding vectors into the last interactive item;
f) embedding the final item into a vector eiAnd carrying out inner product with the recent interest preference vector g of the target user to obtain a prediction score y, sorting according to the prediction score y, and screening the first N items for recommendation.
Further, the value of N in step b) is 10.
Further, step c) comprises the steps of:
c-1) to join sets of session items
Figure BDA0003309413780000031
The item in (1) is used as a graph node to initialize a click adjacency matrix AclickTraversing the click sequences of the target user and the neighbor users, if the item j is the next click item after the user clicks the item i, constructing an edge between the item i and the item j, namely an adjacency matrix AclickMiddle abscissa is i and ordinate is j
Figure BDA0003309413780000032
Increment by 1, for the adjacency matrix AclickThe click conversation graph is obtained through normalization
Figure BDA0003309413780000033
c-2) to federate a set of session items
Figure BDA0003309413780000034
As graph node, initializing the collection adjacency matrix AcollectTraversing the collection sequence of the target user and the neighbor user, if the item j is the next collection item after the user collects the item i, constructing an edge between the item i and the item j, namely an adjacent matrix AcollectMiddle abscissa is i and ordinate is j
Figure BDA0003309413780000035
Increment by 1, for the adjacency matrix AcollectNormalization results in a collection session graph
Figure BDA0003309413780000036
c-3) to federate a set of session items
Figure BDA0003309413780000037
As graph nodes, initialize the purchase adjacency matrix AbuyTraversing the purchasing sequence of the target user and the neighbor user, if the item j is the next purchasing item after the user purchases the item i, constructing an edge between the item i and the item j, namely an adjacent matrix AbuyMiddle abscissa is i and ordinate is j
Figure BDA0003309413780000038
Increment by 1, for the adjacency matrix AbuyThe normalization results in a purchase session graph
Figure BDA0003309413780000039
Further, step d) is performed by the formula
Figure BDA00033094137800000310
Calculating to obtain a final embedded vector e obtained by act behavior high-order propagationi,actAct is one of clicking, collecting and purchasing behaviors,
Figure BDA00033094137800000311
an embedded vector for aggregation of layer I messages of node i in act behavioral session graph, where alIn order to propagate the weights,
Figure BDA00033094137800000312
l is the total number of layers propagated, L2,
Figure BDA00033094137800000313
Figure BDA00033094137800000314
is the item set of the neighbor nodes at the l level of the node i in the act behavior session graph,
Figure BDA00033094137800000315
is ei,actThe weights of node i and neighbor node j in the corresponding adjacency matrix,
Figure BDA00033094137800000316
is composed of
Figure BDA00033094137800000317
The embedded vector of middle node j.
Further, by the formula y ═ softmax (g · e)i) And calculating to obtain a prediction score y, wherein softmax (·) is a normalization function.
The invention has the beneficial effects that: a multi-behavior entitled undirected conversation graph is constructed using the joint multi-behavior sequence data of the target user and the similar users. On the basis, neighbor information is aggregated according to different weights to obtain multi-behavior embedding of the project, multi-behavior embedding of the project is connected in series, and a user interest representation is obtained by combining an attention mechanism. And finally, carrying out inner product by using item embedding and user interest representation to obtain a normalized score to determine whether to recommend the item. Compared with other conversation type recommendation methods, the method has the advantages that the project embedding containing more behavior intentions can be obtained from modeling user multi-behavior sequence data; secondly, the sequence is constructed into a weighted undirected graph, so that the one-way constraint during neighbor information aggregation is removed, and the model can learn the two-way relation among projects; and thirdly, the similar users are used for supplementing the target user data, and the model can learn novel items which do not appear in the target user historical data, so that the diversity and the accuracy of the recommendation result are improved.
Drawings
FIG. 1 is a diagram of a model framework of the present invention.
Detailed Description
The invention is further described below with reference to fig. 1.
A recommendation method based on multi-behavior session graph fusion comprises the following steps:
a) obtaining a target user utClicking, collecting and purchasing historical conversation data and item sets in a certain shopping platform through a browser or a client
Figure BDA0003309413780000041
b) By the formula
Figure BDA0003309413780000042
Computing similarity between users
Figure BDA0003309413780000043
In the formula
Figure BDA0003309413780000044
For using a set of items interacted with by a target user
Figure BDA0003309413780000045
The embedded vector of the initialization is then set to zero,
Figure BDA0003309413780000046
for other users' embedded vectors, all other users are based on similarity
Figure BDA0003309413780000047
The values of the users are sorted in descending order, and the first N users are taken as the neighbor users u according to the sorting resultsObtaining neighbor user usClicking, collecting and storing in a certain shopping platform through a browser or a clientPurchasing historical session data and item sets
Figure BDA0003309413780000051
c) By the formula
Figure BDA0003309413780000052
Calculating to obtain a joint session project set of a target user and a neighbor user
Figure BDA0003309413780000053
To federate a set of session items
Figure BDA0003309413780000054
The item in (1) is a node, and a right undirected click conversation graph is constructed
Figure BDA0003309413780000055
Building collection conversation chart with right and no direction
Figure BDA0003309413780000056
And constructing a purchase session graph with rights and directions
Figure BDA0003309413780000057
d) Click on conversation graph
Figure BDA0003309413780000058
Collection session graph
Figure BDA0003309413780000059
And purchase session directed graph
Figure BDA00033094137800000510
Inputting the three behavior embedded vectors into a LightGCN network to respectively obtain three behavior embedded vectors e of item i aggregation high-order neighbor characteristicsi,click、ei,collectAnd ei,buyBy the formula ei＝ei,click||ei,collect||ei,buyCalculating to obtain a final project embedding vector eiIn the formula, | | is splicingOperating;
e) by the formula
Figure BDA00033094137800000511
Calculating to obtain a recent interest preference vector g of the target user, wherein
Figure BDA00033094137800000512
For the embedded vector of the d-th item i in the current conversation sequence S of the target user,
Figure BDA00033094137800000513
LeakyReLU (-) is a non-linear activation,
Figure BDA00033094137800000514
is a vector of weights, W is a shared parameter matrix,
Figure BDA00033094137800000515
is the embedding vector of the kth item i in the current conversation sequence S of the target user, k belongs to { 1., D }, D is the length of the current conversation sequence S of the target user, e|D|Is a target user utThe items of the last interaction embed the vector.
f) Embedding the final item into a vector eiAnd carrying out inner product with the recent interest preference vector g of the target user to obtain a prediction score y, wherein y represents the probability of the user interacting with the items next, the greater y is the interaction probability, sorting is carried out according to the prediction score y, and the first N items are screened out for recommendation.
A multi-behavior entitled undirected conversation graph is constructed using the joint multi-behavior sequence data of the target user and the similar users. On the basis, neighbor information is aggregated according to different weights to obtain multi-behavior embedding of the project, multi-behavior embedding of the project is connected in series, and a user interest representation is obtained by combining an attention mechanism. And finally, carrying out inner product by using item embedding and user interest representation to obtain a normalized score to determine whether to recommend the item. Compared with other conversation type recommendation methods, the method has the advantages that the project embedding containing more behavior intentions can be obtained from modeling user multi-behavior sequence data; secondly, the sequence is constructed into a weighted undirected graph, so that the one-way constraint during neighbor information aggregation is removed, and the model can learn the two-way relation among projects; and thirdly, the similar users are used for supplementing the target user data, and the model can learn novel items which do not appear in the target user historical data, so that the diversity and the accuracy of the recommendation result are improved.
Example 1:
in order to ensure high connectivity and item diversity of the sparse behavior session graph, it is preferable that the value of N in step b) is 10.
Example 2:
the step c) comprises the following steps:
c-1) to join sets of session items
Figure BDA0003309413780000061
The item in (1) is used as a graph node to initialize a click adjacency matrix AclickTraversing the click sequences of the target user and the neighbor users, if the item j is the next click item after the user clicks the item i, constructing an edge between the item i and the item j, namely an adjacency matrix AclickMiddle abscissa is i and ordinate is j
Figure BDA0003309413780000062
Increment by 1, for the adjacency matrix AclickThe click conversation graph is obtained through normalization
Figure BDA0003309413780000063
c-2) to federate a set of session items
Figure BDA0003309413780000064
As graph node, initializing the collection adjacency matrix AcollectTraversing the collection sequence of the target user and the neighbor user, if the item j is the next collection item after the user collects the item i, constructing an edge between the item i and the item j, namely an adjacent matrix AcollectMiddle abscissa is i and ordinate is j
Figure BDA0003309413780000065
The number of the bits is increased by 1,for adjacent matrix AcollectNormalization results in a collection session graph
Figure BDA0003309413780000066
c-3) to federate a set of session items
Figure BDA0003309413780000067
As graph nodes, initialize the purchase adjacency matrix AbuyTraversing the purchasing sequence of the target user and the neighbor user, if the item j is the next purchasing item after the user purchases the item i, constructing an edge between the item i and the item j, namely an adjacent matrix AbuyMiddle abscissa is i and ordinate is j
Figure BDA0003309413780000068
Increment by 1, for the adjacency matrix AbuyThe normalization results in a purchase session graph
Figure BDA0003309413780000069
Example 3:
in step d) by the formula
Figure BDA0003309413780000071
Calculating to obtain a final embedded vector e obtained by act behavior high-order propagationi,actAct is one of clicking, collecting and purchasing behaviors,
Figure BDA0003309413780000072
an embedded vector for aggregation of layer I messages of node i in act behavioral session graph, where alIn order to propagate the weights,
Figure BDA0003309413780000073
l is the total number of layers propagated, L2,
Figure BDA0003309413780000074
Figure BDA0003309413780000075
is the item set of the neighbor nodes at the l level of the node i in the act behavior session graph,
Figure BDA0003309413780000076
is ei,actThe weights of node i and neighbor node j in the corresponding adjacency matrix,
Figure BDA0003309413780000077
is composed of
Figure BDA0003309413780000078
The embedded vector of middle node j.
Example 4:
by the formula y ═ softmax (g · e)i) And calculating to obtain a prediction score y, wherein softmax (·) is a normalization function.
Finally, it should be noted that: although the present invention has been described in detail with reference to the foregoing embodiments, it will be apparent to those skilled in the art that changes may be made in the embodiments and/or equivalents thereof without departing from the spirit and scope of the invention. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims (5)

1. A recommendation method based on multi-behavior session graph fusion is characterized by comprising the following steps:
a) obtaining a target user utClicking, collecting and purchasing historical conversation data and item sets in a certain shopping platform through a browser or a client
Figure FDA0003309413770000011
b) By the formula
Figure FDA0003309413770000012
Computing similarity between users
Figure FDA0003309413770000013
In the formula
Figure FDA0003309413770000014
For using a set of items interacted with by a target user
Figure FDA0003309413770000015
The embedded vector of the initialization is then set to zero,
Figure FDA0003309413770000016
for other users' embedded vectors, all other users are based on similarity
Figure FDA0003309413770000017
The values of the users are sorted in descending order, and the first N users are taken as the neighbor users u according to the sorting resultsObtaining neighbor user usClicking, collecting and purchasing historical conversation data and item sets in a certain shopping platform through a browser or a client
Figure FDA0003309413770000018
c) By the formula
Figure FDA0003309413770000019
Calculating to obtain a joint session project set of a target user and a neighbor user
Figure FDA00033094137700000110
To federate a set of session items
Figure FDA00033094137700000111
The item in (1) is a node, and a right undirected click conversation graph is constructed
Figure FDA00033094137700000112
Building collection conversation chart with right and no direction
Figure FDA00033094137700000113
And constructing a purchase session graph with rights and directions
Figure FDA00033094137700000114
d) Click on conversation graph
Figure FDA00033094137700000115
Collection session graph
Figure FDA00033094137700000116
And purchase session directed graph
Figure FDA00033094137700000117
Inputting the three behavior embedded vectors into a LightGCN network to respectively obtain three behavior embedded vectors e of item i aggregation high-order neighbor characteristicsi,click、ei,collectAnd ei,buyBy the formula ei＝ei,click||ei,collect||ei,buyCalculating to obtain a final project embedding vector eiIn the formula, | | is splicing operation;
e) by the formula
Figure FDA00033094137700000118
Calculating to obtain a recent interest preference vector g of the target user, wherein
Figure FDA00033094137700000119
For the embedded vector of the d-th item i in the current conversation sequence S of the target user,
Figure FDA0003309413770000021
LeakyReLU (-) is a non-linear activation,
Figure FDA0003309413770000022
is a vector of weights, W is a shared parameter matrix,
Figure FDA0003309413770000023
is the embedding vector of the kth item i in the current conversation sequence S of the target user, k belongs to { 1., D }, D is the length of the current conversation sequence S of the target user, e|D|Is a target user utEmbedding vectors into the last interactive item;
f) embedding the final item into a vector eiAnd carrying out inner product with the recent interest preference vector g of the target user to obtain a prediction score y, sorting according to the prediction score y, and screening the first N items for recommendation.
2. The recommendation method based on multi-behavior session graph fusion as claimed in claim 1, wherein: the value of N in the step b) is 10.
3. The recommendation method based on multi-behavior session graph fusion according to claim 1, wherein the step c) comprises the following steps:
c-1) to join sets of session items
Figure FDA0003309413770000024
The item in (1) is used as a graph node to initialize a click adjacency matrix AclickTraversing the click sequences of the target user and the neighbor users, if the item j is the next click item after the user clicks the item i, constructing an edge between the item i and the item j, namely an adjacency matrix AclickMiddle abscissa is i and ordinate is j
Figure FDA0003309413770000025
Increment by 1, for the adjacency matrix AclickThe click conversation graph is obtained through normalization
Figure FDA0003309413770000026
c-2) to federate a set of session items
Figure FDA0003309413770000027
In (1)Item as graph node initialization collection adjacency matrix AcollectTraversing the collection sequence of the target user and the neighbor user, if the item j is the next collection item after the user collects the item i, constructing an edge between the item i and the item j, namely an adjacent matrix AcollectMiddle abscissa is i and ordinate is j
Figure FDA0003309413770000028
Increment by 1, for the adjacency matrix AcollectNormalization results in a collection session graph
Figure FDA0003309413770000029
c-3) to federate a set of session items
Figure FDA00033094137700000210
As graph nodes, initialize the purchase adjacency matrix AbuyTraversing the purchasing sequence of the target user and the neighbor user, if the item j is the next purchasing item after the user purchases the item i, constructing an edge between the item i and the item j, namely an adjacent matrix AbuyMiddle abscissa is i and ordinate is j
Figure FDA0003309413770000031
Increment by 1, for the adjacency matrix AbuyThe normalization results in a purchase session graph
Figure FDA0003309413770000032
4. The recommendation method based on multi-behavior session graph fusion as claimed in claim 1, wherein: in step d) by the formula
Figure FDA0003309413770000033
Calculating to obtain a final embedded vector e obtained by act behavior high-order propagationi,actAct is one of clicking, collecting and purchasing behaviors,
Figure FDA0003309413770000034
an embedded vector for aggregation of layer I messages of node i in act behavioral session graph, where alIn order to propagate the weights,
Figure FDA0003309413770000035
l is the total number of layers propagated, L2,
Figure FDA0003309413770000036
Figure FDA0003309413770000037
is the item set of the neighbor nodes at the l level of the node i in the act behavior session graph,
Figure FDA0003309413770000038
is ei,actThe weights of node i and neighbor node j in the corresponding adjacency matrix,
Figure FDA0003309413770000039
is composed of
Figure FDA00033094137700000310
The embedded vector of middle node j.
5. The recommendation method based on multi-behavior session graph fusion as claimed in claim 1, wherein: by the formula y ═ softmax (g · e)i) And calculating to obtain a prediction score y, wherein softmax (·) is a normalization function.
CN202111212853.8A 2021-10-19 2021-10-19 Recommendation method based on multi-behavior session graph fusion Active CN113868537B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111212853.8A CN113868537B (en) 2021-10-19 2021-10-19 Recommendation method based on multi-behavior session graph fusion

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111212853.8A CN113868537B (en) 2021-10-19 2021-10-19 Recommendation method based on multi-behavior session graph fusion

Publications (2)

Publication Number Publication Date
CN113868537A true CN113868537A (en) 2021-12-31
CN113868537B CN113868537B (en) 2022-07-05

Family

ID=79000225

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111212853.8A Active CN113868537B (en) 2021-10-19 2021-10-19 Recommendation method based on multi-behavior session graph fusion

Country Status (1)

Country Link
CN (1) CN113868537B (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115329211A (en) * 2022-08-01 2022-11-11 山东省计算中心（国家超级计算济南中心） Personalized interest recommendation method based on self-supervision learning and graph neural network
CN116595157A (en) * 2023-07-17 2023-08-15 江西财经大学 Dynamic interest transfer type session recommendation method and system based on user intention fusion

Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20190362408A1 (en) * 2018-05-25 2019-11-28 Target Brands, Inc. Personalized recommendations for unidentified users based on web browsing context
CN112115352A (en) * 2020-08-28 2020-12-22 齐鲁工业大学 Session recommendation method and system based on user interests
CN112148975A (en) * 2020-09-21 2020-12-29 北京百度网讯科技有限公司 Session recommendation method, device and equipment
CN112364976A (en) * 2020-10-14 2021-02-12 南开大学 User preference prediction method based on session recommendation system
WO2021038592A2 (en) * 2019-08-30 2021-03-04 Tata Consultancy Services Limited System and method for handling popularity bias in item recommendations
CN112733018A (en) * 2020-12-31 2021-04-30 哈尔滨工程大学 Session recommendation method based on graph neural network GNN and multi-task learning
WO2021114590A1 (en) * 2019-12-09 2021-06-17 北京百度网讯科技有限公司 Session recommendation method and apparatus, and electronic device
CN113434756A (en) * 2021-06-18 2021-09-24 山东省人工智能研究院 Personalized comment recommendation method based on graph bidirectional aggregation network link prediction model
CN113487018A (en) * 2021-07-28 2021-10-08 辽宁工程技术大学 Global context enhancement graph neural network method based on session recommendation

Patent Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20190362408A1 (en) * 2018-05-25 2019-11-28 Target Brands, Inc. Personalized recommendations for unidentified users based on web browsing context
WO2021038592A2 (en) * 2019-08-30 2021-03-04 Tata Consultancy Services Limited System and method for handling popularity bias in item recommendations
WO2021114590A1 (en) * 2019-12-09 2021-06-17 北京百度网讯科技有限公司 Session recommendation method and apparatus, and electronic device
CN112115352A (en) * 2020-08-28 2020-12-22 齐鲁工业大学 Session recommendation method and system based on user interests
CN112148975A (en) * 2020-09-21 2020-12-29 北京百度网讯科技有限公司 Session recommendation method, device and equipment
CN112364976A (en) * 2020-10-14 2021-02-12 南开大学 User preference prediction method based on session recommendation system
CN112733018A (en) * 2020-12-31 2021-04-30 哈尔滨工程大学 Session recommendation method based on graph neural network GNN and multi-task learning
CN113434756A (en) * 2021-06-18 2021-09-24 山东省人工智能研究院 Personalized comment recommendation method based on graph bidirectional aggregation network link prediction model
CN113487018A (en) * 2021-07-28 2021-10-08 辽宁工程技术大学 Global context enhancement graph neural network method based on session recommendation

Non-Patent Citations (8)

* Cited by examiner, † Cited by third party
Title
MINGGE ZHANG ET AL.: "GACOforRec_ Session-Based Graph Convolutional Neural Networks Recommendation Model", 《DIGITALOBJECTIDENTIFIER10.1109/ACCESS.2019.2936461》 *
MINGGE ZHANG ET AL.: "GACOforRec_ Session-Based Graph Convolutional Neural Networks Recommendation Model", 《DIGITALOBJECTIDENTIFIER10.1109/ACCESS.2019.2936461》, vol. 7, 29 August 2019 (2019-08-29), pages 114077 - 114085, XP011742489, DOI: 10.1109/ACCESS.2019.2936461 *
SHU WU ET AL.: "Session-based Recommendation with Graph Neural Networks", 《ARXIV:1811.00855V3》 *
SHU WU ET AL.: "Session-based Recommendation with Graph Neural Networks", 《ARXIV:1811.00855V3》, 15 November 2018 (2018-11-15), pages 1 - 9 *
XIN XIA ET AL.: "Self-Supervised Hypergraph Session-based Recommendation Convolutional Networks for", 《ARXIV:2012.06852V4》 *
XIN XIA ET AL.: "Self-Supervised Hypergraph Session-based Recommendation Convolutional Networks for", 《ARXIV:2012.06852V4》, 15 March 2021 (2021-03-15), pages 1 - 9 *
胡承佐等: "基于复杂结构信息的图神经网络序列推荐算法", 《计算机工程》 *
胡承佐等: "基于复杂结构信息的图神经网络序列推荐算法", 《计算机工程》, 9 June 2021 (2021-06-09), pages 1 - 14 *

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115329211A (en) * 2022-08-01 2022-11-11 山东省计算中心（国家超级计算济南中心） Personalized interest recommendation method based on self-supervision learning and graph neural network
CN115329211B (en) * 2022-08-01 2023-06-06 山东省计算中心（国家超级计算济南中心） Personalized interest recommendation method based on self-supervision learning and graph neural network
CN116595157A (en) * 2023-07-17 2023-08-15 江西财经大学 Dynamic interest transfer type session recommendation method and system based on user intention fusion
CN116595157B (en) * 2023-07-17 2023-09-19 江西财经大学 Dynamic interest transfer type session recommendation method and system based on user intention fusion

Also Published As

Publication number Publication date
CN113868537B (en) 2022-07-05

Similar Documents

Publication Publication Date Title
CN111428147B (en) Social recommendation method of heterogeneous graph volume network combining social and interest information
CN112364976B (en) User preference prediction method based on session recommendation system
CN111160954B (en) Recommendation method facing group object based on graph convolution network model
CN112084427A (en) Interest point recommendation method based on graph neural network
CN111222332A (en) Commodity recommendation method combining attention network and user emotion
CN113868537B (en) Recommendation method based on multi-behavior session graph fusion
CN114817663B (en) Service modeling and recommendation method based on class perception graph neural network
CN112115377A (en) Graph neural network link prediction recommendation method based on social relationship
CN109062962A (en) A kind of gating cycle neural network point of interest recommended method merging Weather information
CN110489661B (en) Social relationship prediction method based on generation of confrontation network and transfer learning
CN113641920A (en) Commodity personalized recommendation method and system based on community discovery and graph neural network
CN112685504A (en) Production process-oriented distributed migration chart learning method
CN113379494A (en) Commodity recommendation method and device based on heterogeneous social relationship and electronic equipment
CN112667920A (en) Text perception-based social influence prediction method, device and equipment
CN114036405A (en) Social contact recommendation method and system based on graph convolution network
CN112364245A (en) Top-K movie recommendation method based on heterogeneous information network embedding
CN115203550A (en) Social recommendation method and system for enhancing neighbor relation
CN114528490A (en) Self-supervision sequence recommendation method based on long-term and short-term interests of user
CN118071400A (en) Application method and system based on graph computing technology in information consumption field
CN116701781A (en) Article homography neighborhood aggregation recommendation algorithm based on graph convolution neural network
CN116306834A (en) Link prediction method based on global path perception graph neural network model
CN115935079A (en) Graph collaborative filtering recommendation method based on clusters
CN115577757A (en) Community discovery method based on graph convolution network node label heteromorphic graph
CN114387007A (en) Acceptance enhancement-based dynamic recommendation method for graph neural network
CN114943019A (en) Top k non-overlapping diversified community discovery method based on double-layer weight network random walk

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant