CN-122019868-A - Associated user network data mining method based on multi-information fusion

CN122019868ACN 122019868 ACN122019868 ACN 122019868ACN-122019868-A

Abstract

The invention discloses a related user network data mining method based on multi-information fusion, which comprises the steps of outputting user interaction coefficients through a user interaction model, obtaining article interaction coefficients through an article interaction model, obtaining article activity coefficients through an article activity model, obtaining social trust coefficients through a social trust model, constructing a user-article interaction model to output the user-article interaction coefficients, comparing the obtained user-article interaction coefficients of users with the user-article interaction coefficients of articles which are not contacted yet, obtaining articles higher than the user-article interaction coefficients of the users, obtaining the user-article interaction coefficients of all articles, sequencing the articles from top to bottom, and selecting the first N articles as recommendation results to output. According to the invention, the quantitative result of multi-source fusion is converted into accurate personalized recommendation based on the sorting and screening mechanism of the interaction coefficient, so that the problem of insufficient recommendation adaptability caused by multi-factor splitting in the prior art is effectively solved.

Inventors

Kang Xiancai
Hua Chuangli
YAO YELI

Assignees

浙江广厦建设职业技术大学

Dates

Publication Date: 20260512
Application Date: 20260105

Claims (10)

1. The related user network data mining method based on multi-information fusion is characterized by comprising the following steps of: S1, outputting a user interaction coefficient through a user interaction model based on user interaction data of a user u on an object i; s2, acquiring an article interaction coefficient through an article interaction model based on article interaction data of the article i; s3, acquiring an article activity coefficient through an article activity model based on article activity data of the article i; S4, based on user social trust data of the user u on the object i, acquiring a social trust coefficient through a social trust model; S5, constructing a user-object interaction model based on the social trust coefficient, the user interaction coefficient under the object state coefficient and the object interaction coefficient, and outputting the user-object interaction coefficient; S6, comparing the obtained user-article interaction coefficient of the user with the user-article interaction coefficient of the article which is not contacted yet, and obtaining an article which is higher than the user-article interaction coefficient of the user; and S7, for the user u, acquiring user-article interaction coefficients of the user u for all articles, sorting the articles from high to low, and selecting the first N articles as recommendation results to output.
2. The method for mining data of an associated user network based on multi-information fusion as set forth in claim 1, wherein in S1, the step of outputting the user interaction coefficient through the user interaction model based on the user interaction data of the user u to the object i is: S11, acquiring user interaction data, wherein the user interaction data comprise a user grading value average value, the number of times of accessing the object page and the access time length of the object page; s12, carrying out maximum-minimum normalization processing on the user grading value average value, the access times of the object pages and the access time of the object pages to obtain a user grading value index, a page access times index and a page access time index; And S13, importing the user scoring index, the page access time index and the page access duration index into a user interaction model to obtain a user interaction coefficient.
3. The method for mining the associated user network data based on the multi-information fusion according to claim 2, wherein the user interaction model obtains a user interaction coefficient through a weighting function calculation, and the user interaction coefficient is positively related to the individual interaction intensity of the user on the object; the user interaction model is expressed as: ; Wherein, the Representing the coefficients of the user interaction and, Representing the user's score index, An index indicating the number of page accesses, Represents the page access duration index, Represents a weight coefficient, an , And the larger the value the stronger the user's interaction with the item.
4. The method for mining the associated user network data based on the multi-information fusion according to claim 1, wherein in the step S2, the step of obtaining the item interaction coefficient through the item interaction model based on the item interaction data of the item i is as follows: S21, acquiring item interaction data, wherein the item interaction data comprise the total grading times of the items, the grading mean value of the items and the total browsing amount of the items; S22, summing the total number of scores of the articles, the average value of scores of the articles, the total browsing of the articles, the average value of scores of all the articles and the average value of browsing amounts of all the articles, and obtaining a score number reference value, a score reference value and a browsing amount reference value; S23, carrying out ratio processing on the total scoring times, the scoring mean value and the total browsing quantity of the articles and corresponding reference values to obtain scoring times indexes, scoring mean value indexes and browsing quantity indexes; and S24, importing the scoring times index, the scoring mean index and the browsing quantity index into the article interaction model to obtain the article interaction coefficient.
5. The method for mining the associated user network data based on the multi-information fusion of claim 4, wherein the item interaction model calculates an item interaction coefficient through an aggregation function, and the item interaction coefficient is positively related to the popularity of the item; The item interaction model is expressed as: ; Wherein, the Representing the coefficient of interaction of the article, The index of the number of scores is represented, Represents the score mean value index, The index of the browsing amount is represented, And the larger the value the more popular the item.
6. The method for mining the associated user network data based on the multi-information fusion according to claim 1, wherein in the step S3, the step of acquiring the article activity coefficient through the article activity model based on the article activity data of the article i is as follows: s31, acquiring article interaction data, including article shelf time, article life cycle and article forwarding times; s32, carrying out ratio processing on the article shelf time and the article life cycle to obtain an article aging index; S33, carrying out summation treatment on the article forwarding times and the average value of all the article forwarding times to obtain a forwarding reference value, and carrying out ratio treatment on the article forwarding times and the article reference value to obtain an article forwarding times index; And S34, importing the article aging index and the article forwarding frequency index into an article activity model to obtain an article activity coefficient.
7. The method for mining the associated user network data based on the multi-information fusion of claim 6, wherein the object activity model calculates object activity coefficients through a weighted combination function, and the object activity coefficients are positively related to object freshness and propagation heat; The item activity model is expressed as: ; Wherein, the Representing the activity coefficient of the article, Indicating the age index of the article, An index indicating the number of times the article is forwarded, Representing the balance weight coefficient and , And the greater the value the more fresh and active the item.
8. The method for mining data of an associated user network based on multi-information fusion as set forth in claim 1, wherein in the step S4, based on user social trust data of the user u on the object i, the step of obtaining the social trust coefficient through the social trust model is as follows: S41, carrying out maximum-minimum normalization processing on the user forwarding time to obtain a user forwarding time index; s42, acquiring social trust data of the user, wherein the social trust data comprise the sharing times of the articles, the maximum common friend number in the shared friends and the user forwarding time; s43, carrying out summation processing on the sharing times of the objects, the maximum common friend number in the sharing friends and the corresponding average value to obtain a sharing times reference value and a common friend reference value; S44, carrying out ratio processing on the sharing times of the objects, the maximum common friend number in the sharing friends and the corresponding reference value to obtain a sharing times index and a common friend index; s45, importing the user forwarding time index, the sharing frequency index and the common friend index into a social trust model to obtain a social trust coefficient.
9. The method for mining the associated user network data based on multi-information fusion according to claim 8, wherein the social trust coefficient is positively correlated with the sharing frequency index and the common friend index in the social trust model and negatively correlated with the user forwarding time index; the social trust model is expressed as: ; Wherein, the Representing the coefficient of social trust, Representing the index of the forwarding time and, An index indicating the number of shares is presented, Representing the common friend index (co-friend index), And the larger the value the user For articles The greater the social trust.
10. The method for mining data of an associated user network based on multi-information fusion as set forth in claim 1, wherein in S5, the step of constructing a user-object interaction model based on the social trust coefficient, the user interaction coefficient under the object state coefficient, and the object interaction coefficient to output the user-object interaction coefficient is as follows: S51, acquiring a social trust coefficient and an article state coefficient and taking the complement thereof, and acquiring a social trust coefficient factor and an article state coefficient factor; s52, importing the social trust coefficient factor and the article state coefficient factor into a social context enhancement model to obtain a social context enhancement coefficient; s53, constructing a user-object interaction model based on the user interaction coefficient and the object interaction coefficient under the social situation enhancement coefficient to output the user-object interaction coefficient; The social context enhancement model is expressed as: ; Wherein, the Representing the social context enhancement factor, Representing the factor of the social trust coefficient, Representing an item activity factor; the user-item interaction model is expressed as: ; wherein the user-item interaction coefficient is represented, Representing the coefficients of the user interaction and, Representing the coefficient of interaction of the article, Representing the social context enhancement factor, And the greater the value the greater the interest of user u in item i.

Description

Associated user network data mining method based on multi-information fusion Technical Field The invention belongs to the technical field of information fusion, and particularly relates to a related user network data mining method based on multi-information fusion. Background Along with the rapid development of internet technology and the rapid growth of the scale of user data, how to accurately mine the interests of users and conduct personalized recommendation from massive network behaviors has become a core requirement in the fields of electronic commerce, social networks, content platforms and the like. In the personalized recommendation field, in order to alleviate the problem of data sparsity and utilize social information, various schemes are provided in the prior art, for example, china patent application No. 202410893430.4 discloses a lightweight graph convolution collaborative filtering recommendation method fusing high-order social relations, firstly, user, project and friend information are mapped to a low-dimensional dense vector space through a graph embedding technology in a model embedding layer, so that negative influence of the data sparsity on a model recommendation result is relieved, secondly, a topological structure of a user social relation graph is learned through stacking three layers of graph layers in the graph convolution layer, high-order connection information among the user, the project and the friend is learned, a series of indirect feedback is generated from the implicit negative feedback, the implicit negative feedback is indirectly captured through analyzing user behaviors and friend intimacy to improve the utilization rate of the implicit negative feedback, and besides, the contribution value of the neighbors is measured through a graph attention network in a fusion mode, the self-adaptive dynamic allocation weight is used for the model, the model can filter the noise so that the model has robustness, and a certain interpretability is given to the model. The patent discloses a method for mining high-order social relationships of users to make recommendations by constructing a "user-friend social graph" and a "user-project-friend high-order connection graph" and aggregating multi-order neighborhood information using a graph rolling network (GCN). The method introduces social information through a graph structure, but has the following limitations that firstly, the method is characterized in that the topological structure relation of a social network is learned, the social behavior (such as forwarding and sharing trust) is lack of fine-granularity dynamic quantification, secondly, the scheme mainly fuses two types of information of user-project interaction and user-user social interaction, the important dimension of the dynamic state (such as the life cycle of the article and real-time heat change) of the article cannot be systematically considered, and finally, the GNN-based scheme possibly has the challenges of high training complexity and difficult real-time updating when facing a very large-scale dynamic network. Disclosure of Invention The invention aims to provide a related user network data mining method based on multi-information fusion, which aims to solve the problems in the background technology. The related user network data mining method based on multi-information fusion has the characteristics of realizing self-adaptive weighting of multiple factors and remarkably improving scene adaptability and credibility of a recommendation result. In order to achieve the above purpose, the invention provides a related user network data mining method based on multi-information fusion, which comprises the following steps: S1, outputting a user interaction coefficient through a user interaction model based on user interaction data of a user u on an object i; s2, acquiring an article interaction coefficient through an article interaction model based on article interaction data of the article i; s3, acquiring an article activity coefficient through an article activity model based on article activity data of the article i; S4, based on user social trust data of the user u on the object i, acquiring a social trust coefficient through a social trust model; S5, constructing a user-object interaction model based on the social trust coefficient, the user interaction coefficient under the object state coefficient and the object interaction coefficient, and outputting the user-object interaction coefficient; S6, comparing the obtained user-article interaction coefficient of the user with the user-article interaction coefficient of the article which is not contacted yet, and obtaining an article which is higher than the user-article interaction coefficient of the user; and S7, for the user u, acquiring user-article interaction coefficients of the user u for all articles, sorting the articles from high to low, and selecting the first N articles as recommendation results to output. In th