CN112883268A - Session recommendation method considering user multiple interests and social influence - Google Patents

Session recommendation method considering user multiple interests and social influence Download PDF

Info

Publication number
CN112883268A
CN112883268A CN202110197724.XA CN202110197724A CN112883268A CN 112883268 A CN112883268 A CN 112883268A CN 202110197724 A CN202110197724 A CN 202110197724A CN 112883268 A CN112883268 A CN 112883268A
Authority
CN
China
Prior art keywords
user
interest
target user
friends
target
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
CN202110197724.XA
Other languages
Chinese (zh)
Other versions
CN112883268B (en
Inventor
顾盼
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.)
Zhejiang Zhiduo Network Technology Co ltd
Original Assignee
China Jiliang 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 China Jiliang University filed Critical China Jiliang University
Priority to CN202110197724.XA priority Critical patent/CN112883268B/en
Publication of CN112883268A publication Critical patent/CN112883268A/en
Application granted granted Critical
Publication of CN112883268B publication Critical patent/CN112883268B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/95—Retrieval from the web
    • G06F16/953—Querying, e.g. by the use of web search engines
    • G06F16/9535—Search customisation based on user profiles and personalisation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • 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
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/06—Buying, selling or leasing transactions
    • G06Q30/0601—Electronic shopping [e-shopping]
    • G06Q30/0631—Item recommendations

Landscapes

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

Abstract

The invention discloses a conversation recommendation method considering multi-interest and social influence of a user, which predicts the click rate of the user on a target item based on an item sequence in the current conversation of the user and an item sequence in the previous conversation of a friend. The invention combines the multi-interest of the target user with the social influence of friends on the multi-interest of the target user. And when the social influence of the friends on the target user is calculated, the importance of the friends and the difference of the common interests of the friends and the target user are considered. The forward propagation portion of the present invention is mainly composed of four parts: the first part is to obtain the current multi-interest representation of the user according to the item sequence in the current conversation of the target user; the second part is that the main interest representation of the target user friend is obtained according to the social network; the third part is to calculate the social influence of the friend on the target user; and the fourth part is that the click rate of the user on the target item is predicted by combining the social influence of the friends on the target user and the multi-interest characteristics of the user.

Description

Session recommendation method considering user multiple interests and social influence
Technical Field
The invention belongs to the technical field of internet services, and particularly relates to a conversation recommendation method considering multi-interest and social influence of a user.
Background
With the advent and development of the mobile internet era, more and more user behavior data are accumulated on an online platform. Generally, a user finds items of interest from a platform through a search function, but as the number of items increases, it becomes more and more difficult for the user to find suitable items from a large number of items. Therefore, a recommendation system becomes very important, and finds out the most interesting items of the user from the mass data and recommends the items to the user, so that the satisfaction of the user and the commercial value of a company can be greatly improved. Now, even many online business platforms weaken the search function and rely mainly on the recommendation function, thereby greatly reducing the use threshold of users. Such as tremble, today's headlines, etc. Recommendation systems on these platforms typically face the following two challenges.
First, the interests of the user are dynamically changing and diverse. For example, a user may be interested in sporting goods and leisure apparel for one period of time and ornamental goods and breakfast-like foods for another period of time. Second, users often share items with friends on the online platform, and the interests of users tend to be influenced by friends. And the social influence of different friends on the user is different, and the difference is represented by two points: the first point is that the influence degree of the friends on the user is different, some friends are relatively trusted, and some friends are relatively sparsely. The second point is that the common interests of different friends and target users are different. For example, some friends have a common interest in sports and some friends have a common interest in music. The method is a session recommendation method considering the multi-interest and social influence of the user, and simultaneously solves the two challenges.
Existing recommendation methods either ignore that the user's interests are diverse or that social influences are ignored. For example, Ali's Zhou Rui et al consider the user's multiple interests in a click-through rate prediction method (DIEN) based on a deep interest evolution network. The social influence of friends on the user is considered in the dynamic graph attention mechanism-based conversation social recommendation method of Songyeping et al, North China. The method comprises the steps of extracting multiple interests of a user from a current conversation of the user through a multiple interest extraction module, and calculating the influence of friends on the multiple interests of the user through a specific interest social influence extraction module. Here, a session refers to a sequence of items that are user-interacted over a period of time, typically divided by the time interval between user actions. There are also methods that take as a session the interactive behavior for one or more days or weeks.
Disclosure of Invention
The problem of the method is defined as predicting the click rate of the user to the target item based on the item sequence in the current session (session) of the user and the item sequence in the last session of the friend. Here, a session refers to a sequence of items that are user-interacted over a period of time, typically divided by the time interval between user actions. There are also methods that take as a session the interactive behavior for one or more days or weeks. Any session may be denoted as S ═ x1,x2,…,xτ,…,xtIn which xτThe τ th item representing the user interaction. The vector characterization of the session is { x }1,x2,…,xτ,…,xtTherein of
Figure BDA0002946378150000011
d is the length of the item vector representation. The current session of the target user u is STThe last conversation of the kth friend is represented as
Figure BDA0002946378150000012
And the last session of all friends of the target user is represented as
Figure BDA0002946378150000013
Where N (u) is the set of friends of the target user u. At this time, the target item x is recommendednewHas a probability of P (x)new∣ST,SN(u))。
Recommendation systems on online platforms typically face two challenges: first, the interests of the user are dynamically changing and diverse. For example, a user may be interested in sporting goods and leisure apparel for one period of time and ornamental goods and breakfast-like foods for another period of time. Second, users often share items with friends on the online platform, and the interests of users tend to be influenced by friends. And the social influence of different friends on the user is different, and the difference is represented by two points: the first point is that the influence degree of the friends on the user is different, some friends are relatively trusted, and some friends are relatively sparsely. The second point is that the interests of different buddies are different. For example, some friends have a common interest in sports and some friends have a common interest in music. In order to solve the two challenges, the invention adopts the following technical scheme:
a conversation recommendation method considering multi-interest and social influence of users comprises the following steps:
and obtaining the current multi-interest representation of the user according to the item sequence in the current session of the target user. Session S of user current interactionTCan be represented as ST＝{x1,x2,…,xτ,…,xtIn which xτThe τ th item representing user interaction, T denotes the time period index of the current session, and STIs characterized by { x1,x2,…,xτ,…,xtTherein of
Figure BDA0002946378150000014
d is the length of the item vector representation. Extracting the user interest from the article sequence by adopting a capsule network, wherein the pseudo code is as follows:
Figure BDA0002946378150000021
wherein the content of the first and second substances,
Figure BDA0002946378150000022
for the ith item vector characterization in the session,
Figure BDA0002946378150000023
is the mapping matrix for the jth interest. The number parameter of user interests is M. From the item vector x, an interest-specific project (interest-specific project) can be derivediIn which the vector representation under different interest spaces is extracted
Figure BDA0002946378150000024
Figure BDA0002946378150000025
The method is a dynamic routing part in the capsule network, and parameters are input
Figure BDA0002946378150000026
Is the ith item vector representation xiAnd (4) vector characterization mapped to the jth interest space, wherein the input parameter r is the iteration number of the dynamic routing algorithm.
Figure BDA0002946378150000027
Output parameter v of the methodjRepresenting a user multi-interest vector representation. bijIs the connection coefficient of the ith item vector characterization to the jth interest, cijIs a parameter bijNormalized linkage parameter, representing the likelihood that the ith item is the jth interest, and for an item xiThe sum of the possibilities of different interests is 1, i.e. Σjcij1. The square is a common square vector activation function in a capsule network and has the formula
Figure BDA0002946378150000028
And obtaining the main interest representation of the target user friend according to the social network. The last conversation of the kth friend of the target user is represented as
Figure BDA0002946378150000029
Also, in the same manner as above,
Figure BDA00029463781500000210
can be expressed as
Figure BDA00029463781500000211
The vector is characterized as { x1,x2,…,xτ,…,xl}，
Figure BDA00029463781500000212
Namely conversation
Figure BDA00029463781500000213
The number of the articles in (1). Friend's happy occasionInterest is also diversified, only the main interest of friends is concerned in the method, and excessive noise (noise) is prevented from being introduced during information transmission. The method adopts an attention mechanism (attention mechanism) to extract the main interest (main interest) representation of the friend.
Figure BDA00029463781500000214
ατ＝Wασ(Wggfk+Wxxτ)
Figure BDA00029463781500000215
Wherein x isτIs a conversation
Figure BDA00029463781500000216
Chinese item xτThe vector of (a) is characterized,
Figure BDA00029463781500000217
representing a conversation
Figure BDA00029463781500000218
Length of the sequence of items.
Figure BDA00029463781500000219
And
Figure BDA00029463781500000220
is the model training parameter, and σ is the sigmoid function. Alpha is alphaτRepresenting an article xτOf importance, the final result
Figure BDA00029463781500000221
I.e. is a friend fkOf major interest. The module can adaptively focus on more important items, thereby gaining the main interest of friends.
The social influence of the friend on the target user is calculated. Set of friends of target user uAnd n (u), when calculating the social influence of the friends on the target user, not only the importance of different friends but also the influence of different friends on different interests of the target user are considered. Set of friends N (u) interest in target user vjInfluence of fjCan be calculated by the following method:
Figure BDA00029463781500000222
Figure BDA00029463781500000223
votekj＝maxj(akj)·attnkj
Figure BDA00029463781500000224
wherein the content of the first and second substances,
Figure BDA0002946378150000031
is the kth friend f of the target userkCharacterization of major interest, vjIs the jth interest of the target user. a iskjRepresenting the kth friend fkAnd the jth interest of the target user.
Figure BDA0002946378150000032
The social influence on different interests of the target user is different, so that the similarity between different interests of the target user is normalized by a softmax function to obtain a friend fkImpact on target user jth interest attnkj. At this time, Σjattnkj1, friend fkAnd the influence on different interests of the target user has a competitive relationship. And can be further adjusted by a temperature coefficient tau when tau → 0+Friend fkOnly one interest of the target user is affected; and when τ → ∞ is,friend fkThe effects on different interests of the target user only tend to be consistent. attnkjThe importance of the friends is not considered, so that all friends of the target user play a great role in the target user and do not meet the reality. Therefore, use max (a)kj) To embody friend fkFor the importance of the target user, friend fkThe degree of importance depends on the friend fkSimilarity between the primary interests and the best matching ones of the target users. Finally, friend fkThe impact of different interests of the target user is the voteskj＝max(akj)·attnkjAnd the social influence of all friends on the different interests of the target user is fj。
And predicting the click rate of the user on the target item by combining the social influence of the friend on the target user and the multi-interest representation of the user.
Figure BDA0002946378150000033
Figure BDA0002946378150000034
Figure BDA0002946378150000035
Figure BDA0002946378150000036
Wherein v isjJ-th interest representation extracted from the current session of the target user, fjSocial influence on the target user's jth interest for friends.
Figure BDA0002946378150000037
Is a vector join operation. Wherein the content of the first and second substances,
Figure BDA0002946378150000038
and
Figure BDA0002946378150000039
is the parameter that the model needs to be trained, and σ is the sigmoid function. For different target items, the model focuses on different interests of the user.
And constructing a loss function and training model parameters. Predicting value of click rate of target item through user
Figure BDA00029463781500000310
Calculating a predicted value
Figure BDA00029463781500000311
And the true value y, and the error is used to update the model parameters. We use a cross-entropy loss function to guide the update process of model parameters:
Figure BDA00029463781500000312
where y ∈ {0,1} is the true value, representing whether the user clicked on the target item. σ is a sigmoid function. We update the model parameters using Adam optimizer.
The invention has the following beneficial technical effects:
(1) according to the method and the system, more accurate recommendation is performed by combining the multiple interests of the target user and the social influence of the friends on the multiple interests of the target user.
(2) According to the invention, when the social influence of the friends on the target user is calculated, the importance of the friends is considered, and different interests among the friends are considered, so that more detailed modeling is carried out.
Drawings
FIG. 1 is a flowchart illustrating a conversation recommendation method considering user interests and social influence according to the present invention;
FIG. 2 is a model framework diagram of a conversation recommendation method considering user interest and social influence according to the present invention.
Detailed Description
For further understanding of the present invention, the following describes a conversation recommendation method in which the interest of the user and the social influence are considered in detail with reference to specific embodiments, but the present invention is not limited thereto, and those skilled in the art can make insubstantial improvements and adjustments under the core teaching of the present invention, and still fall within the scope of the present invention.
The problem of the method is defined as predicting the click rate of the user to the target item based on the item sequence in the current session (session) of the user and the item sequence in the last session of the friend. Any session may be denoted as S ═ x1,x2,…,xτ,…,xtIn which xτThe τ th item representing the user interaction. The vector characterization of the session is { x }1,x2,…,xτ,…,xtTherein of
Figure BDA00029463781500000313
d is the length of the item vector representation. The current session of the target user u is SSTThe last conversation of the kth friend is represented as
Figure BDA00029463781500000314
And the last session of all friends of the target user is represented as
Figure BDA00029463781500000315
Figure BDA00029463781500000316
Where N (u) is the set of friends of the target user u. At this time, the target item x is recommendednewHas a probability of P (x)new∣S4,SN(u)). The data adopted by the method is a short video public data set of Kuaisou, and comprises click data and non-click data of a user. Wherein the un-clicked data represents that the platform presents the short video to the user, but the user does not click, i.e. negative examples. The click data of the user is a positive sample.
A forward propagation (forward propagation) section of a conversation recommendation method considering user multi-interests and social influence is mainly composed of four sections, as shown in fig. 2. The first part is to obtain the current multi-interest representation of the user according to the item sequence in the current conversation of the target user. The second part is to get the main interest representation of the target user friend according to the social network. The third part is to calculate the social impact of the friend on the target user. And the fourth part is that the click rate of the user on the target item is predicted by combining the social influence of the friends on the target user and the multi-interest characteristics of the user.
As shown in fig. 1, according to one embodiment of the present invention, the method comprises the steps of:
and S100, root of the object sequence in the current conversation of the target user to obtain the current multi-interest representation of the user. Session S of user current interactionTCan be represented as ST＝{x1,x2,…,xτ,…,xtIn which xτThe τ th item representing user interaction, T denotes the time period index of the current session, and STIs characterized by { x1,x2,…,xτ,…,xtTherein of
Figure BDA0002946378150000041
d is the length of the item vector representation. Extracting the user interest from the article sequence by adopting a capsule network, wherein the pseudo code is as follows:
Figure BDA0002946378150000042
wherein the content of the first and second substances,
Figure BDA0002946378150000043
for the ith item vector characterization in the session,
Figure BDA0002946378150000044
is the mapping matrix for the jth interest. The number parameter of the user interest is M, and the value of M in the method is 5. From the item vector x, an interest-specific project (interest-specific project) can be derivediIn which the vector representation under different interest spaces is extracted
Figure BDA0002946378150000045
Figure BDA0002946378150000046
The method is a dynamic routing part in the capsule network, and parameters are input
Figure BDA0002946378150000047
Is the ith item vector representation xiAnd (3) vector representation mapped to the jth interest space, wherein an input parameter r is the iteration number of the dynamic routing algorithm, and the iteration number r is set to be 2 in the experiment.
Figure BDA0002946378150000048
Output parameter v of the methodjRepresenting a user multi-interest vector representation. bijIs the connection coefficient of the ith item vector characterization to the jth interest, cijIs a parameter bijNormalized linkage parameter, representing the likelihood that the ith item is the jth interest, and for an item xiThe sum of the possibilities of different interests is 1, i.e. Σjcij1. The square is a common square vector activation function in a capsule network and has the formula
Figure BDA0002946378150000049
And S200, obtaining the main interest representation of the target user friend according to the social network. The last conversation of the kth friend of the target user is represented as
Figure BDA00029463781500000410
Also, in the same manner as above,
Figure BDA00029463781500000411
can be expressed as
Figure BDA00029463781500000412
The vector is characterized as { x1,x2,…,xτ,…,xl}，
Figure BDA00029463781500000413
Namely conversation
Figure BDA00029463781500000414
The number of the articles in (1). The interests of the friends are also diversified, only the main interests of the friends are concerned in the method, and excessive noise (noise) is prevented from being introduced during information transmission. The method adopts an attention mechanism (attention mechanism) to extract the main interest (main interest) representation of the friend.
Figure BDA00029463781500000415
ατ＝Wασ(Wggfk+Wxxτ)
Figure BDA00029463781500000416
Wherein x isτIs a conversation
Figure BDA0002946378150000051
Chinese item xτThe vector of (a) is characterized,
Figure BDA0002946378150000052
representing a conversation
Figure BDA0002946378150000053
Length of the sequence of items.
Figure BDA0002946378150000054
And
Figure BDA0002946378150000055
is the model training parameter, and σ is the sigmoid function. Alpha is alphaτRepresenting an article xτOf importance, the final result
Figure BDA0002946378150000056
I.e. is a friend fkOf major interest. The module can adaptively focus on more important items, thereby gaining the main interest of friends.
And S300, calculating the social influence of the friends on the target user. The friend set of the target user u is N (u), and when calculating the social influence of the friends on the target user, the importance of different friends and the influence of different friends on different interests of the target user are considered. Set of friends N (u) interest in target user vjInfluence of fjCan be calculated by the following method:
Figure BDA0002946378150000057
Figure BDA0002946378150000058
votekj＝maxj(akj)·attnkj
Figure BDA0002946378150000059
wherein the content of the first and second substances,
Figure BDA00029463781500000510
is the kth friend f of the target userkCharacterization of major interest, vjIs the jth interest of the target user. a iskjRepresenting the kth friend fkAnd the jth interest of the target user.
Figure BDA00029463781500000511
The social influence on different interests of the target user is different, so that the similarity between different interests of the target user is normalized by a softmax function to obtain a friend fkImpact on target user jth interest attnkj. At this time, Σjattnkj1, friend fkAnd the influence on different interests of the target user has a competitive relationship. And can be further adjusted by a temperature coefficient tau when tau → 0+Friend fkOnly one interest of the target user is affected; while when τ → ∞, friend fkThe effects on different interests of the target user only tend to be consistent. attnkjThe importance of the friends is not considered, so that all friends of the target user play a great role in the target user and do not meet the reality. Therefore, use max (a)kj) To embody friend fkFor the importance of the target user, friend fkkThe degree of importance depends on the friend fkkSimilarity between the primary interests and the best matching ones of the target users. Finally, friend fkkThe impact of different interests of the target user is the voteskj＝max(akj)·attnkjAnd the social influence of all friends on the different interests of the target user is fj。
And S400, predicting the click rate of the user on the target item by combining the social influence of the friends on the target user and the multi-interest representation of the user.
Figure BDA00029463781500000512
Figure BDA00029463781500000513
Figure BDA00029463781500000514
Figure BDA00029463781500000515
Wherein v isjJ-th interest representation extracted from the current session of the target user, fjSocial influence on the target user's jth interest for friends.
Figure BDA00029463781500000516
Is a vector join operation. Wherein the content of the first and second substances,
Figure BDA00029463781500000517
and
Figure BDA00029463781500000518
is the parameter that the model needs to be trained, and σ is the sigmoid function. For different target items, the model focuses on different interests of the user.
S500, constructing a loss function and training model parameters. Predicting value of click rate of target item through user
Figure BDA00029463781500000519
Calculating a predicted value
Figure BDA00029463781500000520
And the true value y, and the error is used to update the model parameters. We use a cross-entropy loss function to guide the update process of model parameters:
Figure BDA00029463781500000521
where y ∈ {0,1} is the true value, representing whether the user clicked on the target item. σ is a sigmoid function. We update the model parameters using Adam optimizer.
The foregoing description of the embodiments is provided to facilitate understanding and application of the invention by those skilled in the art. It will be readily apparent to those skilled in the art that various modifications to the above-described embodiments may be made, and the generic principles defined herein may be applied to other embodiments without the use of inventive faculty. Therefore, the present invention is not limited to the above embodiments, and those skilled in the art should make improvements and modifications to the present invention based on the disclosure of the present invention within the protection scope of the present invention.

Claims (2)

1. A conversation recommendation method considering multi-interest and social influence of a user is characterized by comprising the following steps:
obtaining a current multi-interest representation of a user according to an article sequence in a current conversation of a target user; session S of user current interactionTCan be represented as ST＝{x1，x2，…，xτ，…，xtIn which xτThe τ th item representing user interaction, T denotes the time period index of the current session, and STIs characterized by { x1，x2，…，xτ，…，xtTherein of
Figure FDA0002946378140000011
d is the length of the item vector representation; extracting the multiple interests of the user from the article sequence by adopting a capsule network;
obtaining main interest representations of friends of a target user according to a social network; the last conversation of the kth friend of the target user is represented as
Figure FDA0002946378140000012
Also, in the same manner as above,
Figure FDA0002946378140000013
can be expressed as
Figure FDA0002946378140000014
The vector is characterized as { x1，x2，…，xτ，…，xl}，
Figure FDA0002946378140000015
Namely conversation
Figure FDA0002946378140000016
The number of the articles in the bag; the method adopts an attention mechanism (attention mechanism) to extract main interest (main interest) representation of friends, and the formula is as follows:
Figure FDA0002946378140000017
Figure FDA0002946378140000018
Figure FDA0002946378140000019
wherein x isτIs a conversation
Figure FDA00029463781400000110
Chinese item xτThe vector of (a) is characterized,
Figure FDA00029463781400000111
representing a conversation
Figure FDA00029463781400000112
The length of the sequence of items;
Figure FDA00029463781400000113
and
Figure FDA00029463781400000114
is a model training parameter, σ is a sigmoid function; alpha is alphaτRepresenting an article xτOf importance, the final result
Figure FDA00029463781400000115
I.e. is a friend fkOf major interest in; the module can adaptively pay attention to more important articles, so as to obtain the main interest of friends;
calculating the social influence of the friends on the target user; the friend set of the target user u is N (u), and when calculating the social influence of the friends on the target user, the importance of different friends and different social influences are consideredThe influence of friends on different interests of the target user; set of friends N (u) interest in target user vjInfluence of fjCan be calculated by the following method:
Figure FDA00029463781400000116
Figure FDA00029463781400000117
votekj＝maxj(akj)·attnkj
Figure FDA00029463781400000118
wherein the content of the first and second substances,
Figure FDA00029463781400000119
is the kth friend f of the target userkCharacterization of major interest, vjIs the jth interest of the target user; a iskjRepresenting the kth friend fkThe similarity between the main interest of (a) and the jth interest of the target user;
Figure FDA00029463781400000120
is the main single interest of friends, and the social influence on different interests of the target user should be differentiated, so the similarity of different interests of the target user is normalized by the softmax function to obtain the f of friendskImpact on target user jth interest attnkj(ii) a At this time, Σjattnkj1, friend fkThe influence on different interests of the target user has a competitive relationship; and can be further adjusted by a temperature coefficient tau when tau → 0+Friend fkOnly one interest of the target user is affected; while when τ → ∞, friend fkOnly to the target userThe effects of the same interests tend to be consistent; attnkjThe importance of the friends is not considered, so that all the friends of the target user play a great role in the target user and do not accord with the fact; therefore, use max (a)kj) To embody friend fkFor the importance of the target user, friend fkThe degree of importance depends on the friend fkSimilarity between the primary interests and the best matching interests among the target users; finally, friend fkThe impact of different interests of the target user is the voteskj＝max(akj)·attnkjAnd the social influence of all friends on the different interests of the target user is fj；
Predicting the click rate of the user on the target item by combining the social influence of the friends on the target user and the multi-interest representation of the user;
Figure FDA00029463781400000121
Figure FDA00029463781400000122
Figure FDA00029463781400000123
Figure FDA00029463781400000124
wherein v isjJ-th interest representation extracted from the current session of the target user, fjSocial influence for friends on the target user's jth interest;
Figure FDA00029463781400000125
is a vector join operation; wherein the content of the first and second substances,
Figure FDA00029463781400000126
and
Figure FDA00029463781400000127
is the parameter that the model needs to be trained, and σ is the sigmoid function; for different target items, the model focuses on different interests of the user;
constructing a loss function and training model parameters; predicting value of click rate of target item through user
Figure FDA0002946378140000021
Calculating a predicted value
Figure FDA0002946378140000022
And the true value y, and then using the error to update the model parameters; we use a cross-entropy loss function to guide the update process of model parameters:
Figure FDA0002946378140000023
wherein y is the true value and represents whether the user clicks the target object or not; σ is a sigmoid function; we update the model parameters using Adam optimizer.
2. A conversation recommendation method taking into account user multiple interests and social influences as claimed in claim 1, wherein said pseudo code of a capsule network extracting user multiple interests from the sequence of items is:
Figure FDA0002946378140000024
wherein the content of the first and second substances,
Figure FDA0002946378140000025
for the ith item vector characterization in the session,
Figure FDA0002946378140000026
is the mapping matrix of the jth interest; the number parameter of the user interests is M; from the item vector x, an interest-specific project (interest-specific project) can be derivediIn which the vector representation under different interest spaces is extracted
Figure FDA0002946378140000027
Figure FDA0002946378140000028
The method is a dynamic routing part in the capsule network, and parameters are input
Figure FDA0002946378140000029
Is the ith item vector representation xiThe vector representation mapped to the jth interest space, and the input parameter r is the iteration number of the dynamic routing algorithm;
Figure FDA00029463781400000210
output parameter v of the methodjRepresenting a user multi-interest vector representation; bijIs the connection coefficient of the ith item vector characterization to the jth interest, cijIs a parameter bijNormalized linkage parameter, representing the likelihood that the ith item is the jth interest, and for an item xiThe sum of the possibilities of different interests is 1, i.e. Σjcij1 is ═ 1; the square is a common square vector activation function in a capsule network and has the formula
Figure FDA00029463781400000211
CN202110197724.XA 2021-02-22 2021-02-22 Session recommendation method considering user multiple interests and social influence Active CN112883268B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110197724.XA CN112883268B (en) 2021-02-22 2021-02-22 Session recommendation method considering user multiple interests and social influence

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110197724.XA CN112883268B (en) 2021-02-22 2021-02-22 Session recommendation method considering user multiple interests and social influence

Publications (2)

Publication Number Publication Date
CN112883268A true CN112883268A (en) 2021-06-01
CN112883268B CN112883268B (en) 2022-02-01

Family

ID=76056742

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110197724.XA Active CN112883268B (en) 2021-02-22 2021-02-22 Session recommendation method considering user multiple interests and social influence

Country Status (1)

Country Link
CN (1) CN112883268B (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113704441A (en) * 2021-09-06 2021-11-26 中国计量大学 Conversation recommendation method considering importance of item and item attribute feature level
CN114756762A (en) * 2022-06-13 2022-07-15 腾讯科技（深圳）有限公司 Data processing method, device, equipment, storage medium and program product

Citations (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20130080526A1 (en) * 2010-01-11 2013-03-28 Barjinderpal S. Gill Apparatus and method for delivering target content to members on a social network
US20140019453A1 (en) * 2012-07-13 2014-01-16 Telefonaktiebolaget L M Ericsson (Publ) Apparatuses and Methods for Assessing User Interest Scores as Altered by Friends Influence
US20170109709A1 (en) * 2015-09-09 2017-04-20 Steven Wu Computerized systems and methods for offline activity management
WO2017157149A1 (en) * 2016-03-15 2017-09-21 平安科技（深圳）有限公司 Social network-based recommendation method and apparatus, server and storage medium
CN108241619A (en) * 2016-12-23 2018-07-03 西北大学 A kind of recommendation method based on the more interest of user
CN108256093A (en) * 2018-01-29 2018-07-06 华南理工大学 A kind of Collaborative Filtering Recommendation Algorithm based on the more interest of user and interests change
CN109767301A (en) * 2019-01-14 2019-05-17 北京大学 Recommended method and system, computer installation, computer readable storage medium
US20190325863A1 (en) * 2018-04-20 2019-10-24 Facebook, Inc. Assisting Users with Personalized and Contextual Communication Content
CN112307258A (en) * 2020-11-25 2021-02-02 中国计量大学 Short video click rate prediction method based on double-layer capsule network
US20210173884A1 (en) * 2018-11-29 2021-06-10 Tencent Technology (Shenzhen) Company Limited Recommendation method and apparatus, and storage medium

Patent Citations (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20130080526A1 (en) * 2010-01-11 2013-03-28 Barjinderpal S. Gill Apparatus and method for delivering target content to members on a social network
US20140019453A1 (en) * 2012-07-13 2014-01-16 Telefonaktiebolaget L M Ericsson (Publ) Apparatuses and Methods for Assessing User Interest Scores as Altered by Friends Influence
US20170109709A1 (en) * 2015-09-09 2017-04-20 Steven Wu Computerized systems and methods for offline activity management
WO2017157149A1 (en) * 2016-03-15 2017-09-21 平安科技（深圳）有限公司 Social network-based recommendation method and apparatus, server and storage medium
CN108241619A (en) * 2016-12-23 2018-07-03 西北大学 A kind of recommendation method based on the more interest of user
CN108256093A (en) * 2018-01-29 2018-07-06 华南理工大学 A kind of Collaborative Filtering Recommendation Algorithm based on the more interest of user and interests change
US20190325863A1 (en) * 2018-04-20 2019-10-24 Facebook, Inc. Assisting Users with Personalized and Contextual Communication Content
US20210173884A1 (en) * 2018-11-29 2021-06-10 Tencent Technology (Shenzhen) Company Limited Recommendation method and apparatus, and storage medium
CN109767301A (en) * 2019-01-14 2019-05-17 北京大学 Recommended method and system, computer installation, computer readable storage medium
CN112307258A (en) * 2020-11-25 2021-02-02 中国计量大学 Short video click rate prediction method based on double-layer capsule network

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
PAN GU ET AL: "Enhancing session-based social recommendation through item graph", 《NEUROCOMPUTING》 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113704441A (en) * 2021-09-06 2021-11-26 中国计量大学 Conversation recommendation method considering importance of item and item attribute feature level
CN114756762A (en) * 2022-06-13 2022-07-15 腾讯科技（深圳）有限公司 Data processing method, device, equipment, storage medium and program product

Also Published As

Publication number Publication date
CN112883268B (en) 2022-02-01

Similar Documents

Publication Publication Date Title
WO2020228514A1 (en) Content recommendation method and apparatus, and device and storage medium
CN108460619B (en) Method for providing collaborative recommendation model fusing explicit and implicit feedback
CN110796313B (en) Session recommendation method based on weighted graph volume and item attraction model
CN110879864B (en) Context recommendation method based on graph neural network and attention mechanism
CN109241440A (en) It is a kind of based on deep learning towards implicit feedback recommended method
CN112733018B (en) Session recommendation method based on graph neural network GNN and multi-task learning
CN112883268B (en) Session recommendation method considering user multiple interests and social influence
CN114817663B (en) Service modeling and recommendation method based on class perception graph neural network
CN112765480A (en) Information pushing method and device and computer readable storage medium
CN111241394A (en) Data processing method and device, computer readable storage medium and electronic equipment
CN112819575B (en) Session recommendation method considering repeated purchasing behavior
CN113420221B (en) Interpretable recommendation method integrating implicit article preference and explicit feature preference of user
CN111681084A (en) E-commerce platform recommendation method based on social relationship influence factors
CN113761383A (en) Search and recommendation fusion system based on unified user behavior modeling
CN112396492A (en) Conversation recommendation method based on graph attention network and bidirectional long-short term memory network
CN112948707A (en) Collaborative filtering recommendation algorithm for optimizing LFM (Linear frequency modulation) through reinforcement learning
CN112883289B (en) PMF recommendation method based on social trust and tag semantic similarity
CN112905887B (en) Conversation recommendation method based on multi-interest short-term priority model
CN112307258B (en) Short video click rate prediction method based on double-layer capsule network
CN113344648A (en) Advertisement recommendation method and system based on machine learning
CN112231579A (en) Social video recommendation system and method based on implicit community discovery
CN114547466A (en) Information diversity recommendation method based on deep reinforcement learning
CN115293812A (en) E-commerce platform session perception recommendation prediction method based on long-term and short-term interests
CN113569139A (en) Personalized session recommendation method and system
CN113704441B (en) Conversation recommendation method considering importance of item and item attribute feature level

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
TR01 Transfer of patent right
TR01 Transfer of patent right

Effective date of registration: 20230925

Address after: Room 407-10, floor 4, building 2, Haichuang science and technology center, Cangqian street, Yuhang District, Hangzhou City, Zhejiang Province, 311100

Patentee after: Zhejiang Zhiduo Network Technology Co.,Ltd.

Address before: 310018, No. 258, source street, Xiasha Higher Education Park, Hangzhou, Zhejiang

Patentee before: China Jiliang University