基于图神经网络和标签可重叠社区的社会化影视推荐模型  

A Social Movie Recommendation Model Based on Graph Neural Network and Tag Overlapping Community

作　　者：周欢 刘嘉 王欢芳[1] Zhou Huan

机构地区：[1]湖南工业大学商学院,湖南株洲412007

出　　处：《情报理论与实践》2021年第6期164-170,共7页Information Studies:Theory & Application

基　　金：国家自然科学基金青年项目“社会网络环境下基于灰信息和云模型的影视推荐方法研究”(项目编号:71801090);湖南省自然科学基金青年项目“媒介融合背景下基于扩展灰数和云模型的影视推荐方法研究”(项目编号:2018JJ3132);湖南省社会科学成果评审委员会课题“基于犹豫模糊语言的湖南工业旅游资源评价”(项目编号:XSP18YBZ158)的研究成果。

摘　　要：[目的/意义]随着互联网的迅速普及,海量的在线影视作品导致用户难以快速准确地获取到所需内容,如何为用户进行个性化影视推荐成为学者们研究的热点。[方法/过程]首先利用在线影视平台中的用户交互数据以及电影信息数据建立完全三部图,然后构建图神经网络并结合扩散算法计算标签间关联度,基于标签间关联度获取标签可重叠社区,通过计算用户和电影对标签可重叠社区的归属度及用户与电影的匹配度,最终为用户生成推荐列表。[结果/结论]采用hetrec2011-movielens-2k数据集进行实验分析,结果显示所提出算法模型的召回率、精确率和F1值均优于同类算法。因此,通过聚合关联用户的信息能够充分挖掘当前用户的潜在兴趣偏好,提升算法推荐效果,有助于为用户精准生成个性化推荐内容。[Purpose/significance]With the rapid spread of the Internet,it is more difficult for users to obtain the required information quickly and accurately from diverse television and film works.How to make personalized recommendation for users has become a popular research topic.[Method/process]Firstly,by using the user interaction data and movie’s information data on the online video platform,a complete tripartite graph was established.Then,constructing a graph neural network and calculating the association degree between tags by diffusion algorithm.Based on the association degree of tags,the tag overlapping community was obtained.Finally,the recommendation list for users was generated by calculating the attribution degree of users and movies to the overlapping community and the matching degree between users and movies.[Result/conclusion]The hetrec 2011-movielens-2k dataset was used for experimental analysis,and the results showed that the proposed algorithm model had higher recall rate,accuracy rate and F1 value than other similar algorithms.Therefore,by aggregating the information of related users,the potential interest preferences of current users could be fully explored and the performance of recommendation would improve,which provided helps for users to generate high-quality personalized recommended content.

关 键 词：图神经网络 标签可重叠社区 完全三部图 影视推荐 社会化推荐 

分 类 号：TP391.3[自动化与计算机技术—计算机应用技术] TP183[自动化与计算机技术—计算机科学与技术]