WO2020052039A1

WO2020052039A1 - Optimal social recommendation method and device under limited attention

Info

Publication number: WO2020052039A1
Application number: PCT/CN2018/114616
Authority: WO
Inventors: 朱文武; 王鑫
Original assignee: 清华大学
Priority date: 2018-09-13
Filing date: 2018-11-08
Publication date: 2020-03-19
Also published as: CN109410079A; CN109410079B

Abstract

Disclosed are an optimal social recommendation method and device under limited attention. The method comprises the following steps: acquiring a privacy feature vector of a target user, and obtaining, according to the privacy feature vector, at least one friend who satisfies a pre-set social influence to obtain, through learning, the social influence weight of the at least one friend; estimating a comprehensive preference of the target user according to preferences of the target user and the social influence weight of the at least one friend; and making a social recommendation for the target user according to the comprehensive preference of the target user. By means of the method, a social recommendation is made to the user according to the user's own preference and a friend who has the most influence on the user, thereby effectively improving the accuracy of social recommendation.

Description

Method and device for social recommendation under optimal limited attention

Cross-reference to related applications

This application claims the priority of Chinese Patent Application No. “201811067075.6” submitted by Tsinghua University on September 13, 2018, with the invention name “Social Recommendation Method and Apparatus under Optimal Limited Attention”.

Technical field

The invention relates to the technical field of personalized social recommendation, in particular to a method and device for social recommendation under optimal limited attention.

Background technique

The related technology believes that all the friend information of the target user should have a social impact on the target user. However, work in the social sciences has proven that in fact humans have limited attention and cannot receive information (social influence) from all friends indefinitely. Therefore, the existing methods do not take into account that in practice, users will selectively receive social influences from friends instead of holistic acceptance of the social influences of all friends.

Summary of the Invention

The present invention aims to solve at least one of the technical problems in the related technology.

Therefore, an object of the present invention is to propose a social recommendation method with optimal limited attention, which can effectively improve the accuracy of social recommendation recommendation.

Another object of the present invention is to provide a social recommendation application device with optimal limited attention.

In order to achieve the above object, an embodiment of the present invention provides a method for social recommendation under optimal limited attention, including the following steps: obtaining a privacy feature vector of a target user, and obtaining a preset social satisfaction according to the privacy feature vector Influence at least one friend to learn to get the social influence weight of the at least one friend; and estimate the target user ’s social impact weight based on the target user ’s own preferences and the social influence weight of the at least one friend Comprehensive preferences; social recommendation for the target user based on the overall preferences of the target user.

According to the embodiment of the present invention, the social recommendation application method with optimal limited attention is based on the user's own preferences and the social influence weight of friends. Social recommendation is made to the user, and the concept of limited attention is optimally integrated into the social recommendation, and a novel A social recommendation model that is more in line with the user's actual life scenario and has higher recommendation accuracy, thereby effectively improving the accuracy of social recommendation.

In addition, the social recommendation application method under the optimal limited attention according to the above embodiment of the present invention may also have the following additional technical features:

Further, in an embodiment of the present invention, the obtaining the privacy feature vector of the target user further comprises: obtaining the privacy data of the target user, and obtaining the privacy feature vector of the target user according to a matrix decomposition technique.

Further, in an embodiment of the present invention, the target user privacy data includes the target user's own information and the target user's social relationship information. Further, in an embodiment of the present invention, the privacy feature vector of the target user includes K dimensions, wherein the K dimensions represent K interest directions, and the value of each dimension represents the interest direction corresponding to the dimension Degree of preference, K is a positive integer.

Further, in an embodiment of the present invention, the target user includes a cold start user and a central user, wherein the cold start user is a user who does not have any historical interaction records, and the central user has multiple social relationships User.

In order to achieve the above object, an embodiment of another aspect of the present invention proposes a social recommendation device with optimal limited attention, including: an acquisition module for acquiring a privacy feature vector of a target user, and obtaining the privacy feature vector according to the privacy feature vector At least one friend who satisfies a preset social influence to learn to obtain the social influence weight of the at least one friend; an estimation module is configured to according to the target user's own preferences and the social relationship of the at least one friend The influence weight estimates the comprehensive preferences of the target user; a recommendation module is configured to perform social recommendation for the target user according to the comprehensive preferences of the target user.

According to the embodiment of the present invention, the social recommendation device with optimal limited attention performs social recommendation on the user according to the user's own preferences and the friends who have the most influence on the user, and optimally integrates into the social recommendation through the concept of limited attention. A social recommendation model that is more in line with the user's actual life scenario and has higher recommendation accuracy is proposed, thereby effectively improving the accuracy of social recommendation.

In addition, the social recommendation device under the optimal limited attention according to the above embodiment of the present invention may also have the following additional technical features:

Further, in an embodiment of the present invention, the obtaining module is further configured to obtain privacy data of the target user, and obtain a privacy feature vector of the target user according to a matrix decomposition technique.

Additional aspects and advantages of the present invention will be given in part in the following description, part of which will become apparent from the following description, or be learned through the practice of the present invention.

BRIEF DESCRIPTION OF THE DRAWINGS

The above and / or additional aspects and advantages of the present invention will become apparent and easily understood from the following description of the embodiments with reference to the accompanying drawings, in which:

FIG. 1 is a flowchart of a social recommendation method with optimal limited attention according to an embodiment of the present invention; FIG.

FIG. 2 is a schematic structural diagram of a social recommendation device under optimal limited attention according to an embodiment of the present invention.

detailed description

Hereinafter, embodiments of the present invention will be described in detail. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals represent the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the drawings are exemplary and are intended to explain the present invention, but should not be construed as limiting the present invention.

The method and device for social recommendation under optimal limited attention according to an embodiment of the present invention will be described below with reference to the drawings. First, the method for social recommendation under optimal limited attention according to an embodiment of the present invention will be described with reference to the drawings.

FIG. 1 is a flowchart of a social recommendation method with optimal limited attention according to an embodiment of the present invention.

As shown in FIG. 1, the optimal social attention recommendation method includes the following steps:

In step S101, a privacy feature vector of a target user is obtained, and at least one friend who satisfies a preset social influence is obtained according to the privacy feature vector, so as to learn to obtain a social influence weight of at least one friend.

It can be understood that the embodiment of the present invention considers the limited attention of the user, and optimally selects a plurality of the most influential friends for each target user and learns their corresponding weight values. Specifically, for each target user, k friends who have the most influence on him / her are optimally selected and their corresponding weights are learned, and their social influence on the target user is integrated into the social recommendation application. It should be noted that the embodiment of the present invention proves the optimality of the process of selecting k friends and learning their weights through strict mathematical derivation.

Specifically, according to the embodiment of the present invention, for each target user, according to the data, optimally personally find k friends (less than or equal to the number of all friends of the target user) whose friends have the greatest social impact and learn the corresponding ones of the selected friends. Social influence weights to simulate scenarios where users have limited attention.

Further, in an embodiment of the present invention, obtaining the privacy feature vector of the target user further includes: obtaining the privacy data of the target user, and obtaining the privacy feature vector of the target user according to a matrix decomposition technique.

It can be understood that according to the embodiment of the present invention, according to the user hidden feature vector obtained by the matrix decomposition technology, for each target user, individually find the best k friends with the strongest social influence, and learn the corresponding Social influence weight value.

Specifically, the embodiment of the present invention applies a principle similar to the nearest neighbor search, and uses the user's hidden feature vector obtained by matrix decomposition as an input, and optimally finds for each user the k (the value varies with the user) And different) friends and optimally learn the weight of their social influence.

Among them, matrix decomposition refers to decomposing a matrix into a product of two or more matrices. For the user-product matrix (scoring matrix), it is denoted as R _{m × n} . It can be decomposed into a product of two or more matrices, assuming decomposition into two matrices P _{m × k} and Q _{k × n} . In the embodiment of the present invention, the product of the matrices P _{m × k} and Q _{k × n} can be restored. The original matrix R _{m × n} :

Among them, the matrix P _{m × k} represents the relationship between m users and k topics, and each row represents the hidden feature vector of a user i, denoted as U _i ; and the matrix Q _{k × n} represents k topics The relationship with n products, each column of which represents the hidden feature vector of an item j, denoted as V _j .

For each user i and his friend u, iteratively calculate λ _k :

among them

Both L and C are constant. When λ _k <β _{i, k + 1} ,

That is the weight between user i and his friend u.

Further, in an embodiment of the present invention, the target user privacy data includes the target user's own information and the target user's social relationship information.

Further, in an embodiment of the present invention, the privacy feature vector of the target user includes K dimensions, where the K dimensions represent K interest directions, and the value of each dimension represents the preference degree of the interest direction corresponding to the dimension, K is a positive integer.

It can be understood that each user has a k-dimensional hidden feature vector. The k dimensions represent k interest directions. The value of each dimension in the vector represents the user's preference for the direction of interest corresponding to the dimension (such as , The larger the value, the more interested the user is in this direction).

Further, in one embodiment of the present invention, the target users include a cold start user and a central user, wherein the cold start user is a user who does not have any historical interaction records, and the central user is a user who has multiple social relationships.

Specifically, cold-start users specifically refer to users who do not have any historical interaction records with the system (such as scoring items, clicking on item links, commenting on items, putting items in a shopping cart, etc.). Since there is no historical information of users, the traditional collaborative filtering model cannot recommend such users, and it is necessary to use the information of the user's social network to recommend such users (that is, social recommendation).

For "central users" who have a large number of social relationships (such as friends), first, because humans have limited capacity to receive information, they cannot receive all the information from social relationships; second, because too many social relationships provide rich information while providing rich information , Will also introduce interference information. At this time, how to optimally select such social relationships (within the maximum acceptance ability) that can provide them with effective help information and use these social information to assist in learning the user's preferences has considerable significance.

In step S102, the comprehensive preference of the target user is estimated according to the target user's own preference and the social influence weight of at least one friend.

It can be understood that, according to the embodiment of the present invention, the user's comprehensive preference estimate for an item is obtained under the combined effect of the target user's own preferences and the social influence of the best k friends selected for each user in step S101. To train the parameters in the model. Among them, the model refers to a model that recommends items that may be of interest to the target user by analyzing the target user's privacy data, that is, the user's own information and the user's social relationship information.

Specifically, the embodiment of the present invention integrates the social influence of the selected friend into social recommendation and other model parameters for joint optimization to recommend content for the user. That is to say, in the embodiment of the present invention, the social influence of the most influential friends previously found and the social recommendation can be effectively combined through an EM-like algorithm, so that each user's optimal number of friends, k, and the friend's social Impact weights and hidden feature vectors of users and items serve to improve the accuracy of recommendations. Among them, the embodiment of the present invention jointly optimizes the personalization parameter k, the social influence weight value of the selected friend, and the hidden feature vectors of users and items through an iterative algorithm. The model can better simulate real life scenarios, and is stronger and more efficient. Recommended accuracy.

In step S103, a social recommendation is performed for the target user according to the comprehensive preferences of the target user.

It can be understood that, the embodiment of the present invention is based on a limited attention social recommendation system, and recommends items that may be of interest to the user according to the comprehensive preferences of the user. Specifically, the embodiment of the present invention is based on the social recommendation with optimal limited attention, inferring their interests and making accurate recommendations in a manner more in line with the actual life scene of people.

It should be noted that the embodiment of the present invention takes into account the phenomenon that the user's limited attention in actual life is also confirmed in social sciences. The concept of limited attention is optimally integrated into social recommendation, and a social recommendation model that is more in line with the user's actual life scenario and has higher recommendation accuracy is proposed, so as to effectively recommend for cold-start users and when the user has a lot of friends, Optimally select multiple friends who have the greatest social impact on the target user to receive their social information, so as to find the content of interest more efficiently and accurately.

In summary, the embodiment of the present invention considers the practical problem of limited attention in the process of receiving information by users in real life scenarios, and proposes to introduce the concept of limited attention into social recommendation applications, and to optimally The user selects k (k value varies from user to user) the most influential friends and learns their social influence weights at the same time, so that it is more realistic and more in line with people's daily habits and more accurate social recommendations.

According to the social recommendation application method with optimal limited attention provided by the embodiment of the present invention, the user is socially recommended according to the user's own preferences and the social influence weight of a friend, and is optimally integrated into the social recommendation through the concept of limited attention. A social recommendation model that is more in line with the user's actual life scenario and has higher recommendation accuracy is proposed, thereby effectively improving the accuracy of social recommendation.

Next, a social recommendation device with optimal limited attention according to an embodiment of the present invention will be described with reference to the drawings.

As shown in FIG. 2, the social recommendation device 10 with optimal limited attention includes: an acquisition module 100, an estimation module 200, and a recommendation module 300.

The obtaining module 100 is configured to obtain a privacy feature vector of a target user, and obtain at least one friend who satisfies a preset social influence according to the privacy feature vector, so as to learn to obtain a social influence weight of the at least one friend. The estimation module 200 is configured to estimate the overall preferences of the target user according to the target users' own preferences and the social influence weight of at least one friend. The recommendation module 300 is configured to perform social recommendation for the target user according to the comprehensive preferences of the target user. The device 10 according to the embodiment of the present invention performs social recommendation on the user according to the user's own preferences and friends who have the most influence on the user, thereby effectively improving the accuracy of the social recommendation.

Further, in an embodiment of the present invention, the obtaining module 100 is further configured to obtain the privacy data of the target user, and obtain the privacy feature vector of the target user according to a matrix decomposition technique.

Further, in an embodiment of the present invention, the target user privacy data includes the target user's own information and the target user's social relationship information. Further, in an embodiment of the present invention, the privacy feature vector of the target user includes K dimensions, where the K dimensions represent K interest directions, and the value of each dimension represents the preference degree of the interest direction corresponding to the dimension, K is a positive integer.

It should be noted that, the foregoing explanation of the embodiment of the method for social recommendation application under optimal limited attention is also applicable to the device for social recommendation application under optimal limited attention in this embodiment, which is not repeated here.

According to an embodiment of the present invention, the social recommendation device with optimal limited attention performs social recommendation on the user according to the user's own preferences and friends who have the most influence on the user, and integrates the social recommendation optimally through the concept of limited attention. Among them, a social recommendation model that is more in line with the user's actual life scene and has higher recommendation accuracy is proposed, thereby effectively improving the accuracy of social recommendation.

In the description of the present invention, it should be understood that the terms "center", "longitudinal", "transverse", "length", "width", "thickness", "upper", "lower", "front", " Rear "," left "," right "," vertical "," horizontal "," top "," bottom "," inside "," outside "," clockwise "," counterclockwise "," axial ", The azimuth or position relationship indicated by "radial", "circumferential", etc. is based on the azimuth or position relationship shown in the drawings, and is only for the convenience of describing the present invention and simplifying the description, rather than indicating or suggesting the device or element referred It must have a specific orientation and be constructed and operated in a specific orientation, so it cannot be understood as a limitation of the present invention.

In addition, the terms "first" and "second" are used for descriptive purposes only, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In the description of the present invention, the meaning of "a plurality" is at least two, for example, two, three, etc., unless it is specifically and specifically defined otherwise.

In the present invention, the terms "installation", "connected", "connected", "fixed" and other terms shall be understood in a broad sense unless otherwise specified and defined, for example, they may be fixed connections or removable connections , Or integrated; it can be mechanical or electrical; it can be directly connected or indirectly connected through an intermediate medium, it can be the internal connection of the two elements or the interaction between the two elements, unless otherwise specified The limit. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific situations.

In the present invention, unless explicitly stated and defined otherwise, the first feature "on" or "down" of the second feature may be the first and second features in direct contact, or the first and second features indirectly through an intermediate medium. contact. Moreover, the first feature is "above", "above", and "above" the second feature. The first feature is directly above or obliquely above the second feature, or only indicates that the first feature is higher in level than the second feature. The first feature is “below”, “below”, and “below” of the second feature. The first feature may be directly below or obliquely below the second feature, or it may simply indicate that the first feature is less horizontal than the second feature.

In the description of this specification, the description with reference to the terms “one embodiment”, “some embodiments”, “examples”, “specific examples”, or “some examples” and the like means specific features described in conjunction with the embodiments or examples , Structure, material, or characteristic is included in at least one embodiment or example of the present invention. In this specification, the schematic expressions of the above terms are not necessarily directed to the same embodiment or example. Moreover, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples. In addition, without any contradiction, those skilled in the art may combine and combine different embodiments or examples and features of the different embodiments or examples described in this specification.

Although the embodiments of the present invention have been shown and described above, it can be understood that the above embodiments are exemplary and should not be construed as limitations on the present invention. Those skilled in the art can interpret the above within the scope of the present invention. Embodiments are subject to change, modification, substitution, and modification.

Claims

A method of social recommendation under optimal limited attention, which is characterized by including the following steps:

Obtaining a privacy feature vector of a target user, and obtaining at least one friend who satisfies a preset social influence according to the privacy feature vector, so as to learn to obtain the social influence weight of the at least one friend;

Estimating the comprehensive preferences of the target user according to the target user's own preferences and the social influence weight of the at least one friend; and

Perform social recommendation for the target user according to the comprehensive preferences of the target user.
The method of social recommendation under optimal limited attention according to claim 1, wherein the obtaining a privacy feature vector of a target user further comprises:

The privacy data of the target user is acquired, and a privacy feature vector of the target user is obtained according to a matrix decomposition technique.
The method of social recommendation under optimal limited attention according to claim 2, wherein the target user privacy data includes the target user's own information and the target user's social relationship information.
The method of social recommendation under optimal limited attention according to claim 1 or 2, wherein the privacy feature vector of the target user includes K dimensions, wherein the K dimensions represent K interest directions, The value of each dimension represents the degree of preference of the direction of interest corresponding to that dimension, and K is a positive integer.
The method for social recommendation under optimal limited attention according to any one of claims 1-4, wherein the target user comprises a cold-start user and a central user, wherein the cold-start user has no history A user who records interactions, and the central user is a user who has multiple social relationships.
A social recommendation device with optimal limited attention, which includes:

An obtaining module, configured to obtain a privacy feature vector of a target user, and obtain at least one friend who satisfies a preset social influence according to the privacy feature vector, so as to learn to obtain a social influence weight of the at least one friend;

An estimation module, configured to estimate a comprehensive preference of the target user according to the target user's own preferences and the social influence weight of the at least one friend; and

A recommendation module is configured to perform social recommendation for the target user according to the comprehensive preferences of the target user.
The social recommendation device with optimal limited attention according to claim 6, wherein the obtaining module is further configured to obtain privacy data of the target user, and obtain the privacy of the target user according to a matrix decomposition technique Feature vector.
The social recommendation device with optimal limited attention according to claim 7, wherein the target user privacy data includes the target user's own information and the target user's social relationship information.
The social recommendation device under optimal limited attention according to claim 6 or 7, wherein the privacy feature vector of the target user includes K dimensions, wherein the K dimensions represent K directions of interest, The value of each dimension represents the degree of preference of the direction of interest corresponding to that dimension, and K is a positive integer.
The social recommendation device with optimal limited attention according to any one of claims 6-9, wherein the target user includes a cold start user and a central user, wherein the cold start user has no history A user who records interactions, and the central user is a user who has multiple social relationships.