Research On Tag Recommendation Based On LDA Topic Model In Social Tagging System

Web2.0 has been widely followed since it was put forward in 2004.It focuses on the concept of human-computer interaction centered on "users",so that users are no longer in a passive stage,they have mastered the initiative and can communicate with the Web anytime and anywhere.They are not only recipients of information,but also the creators and publishers of information.In this open Internet environment,everyone can participate in,create and share information.Social tagging is an emerging method of organizing network information resources in the Web2.0 era,users can use keywords(Tag)to tag various resources in the network according to their own wishes,to organize,classify and retrieve resources more effectively.Social tagging now has been widely used,but the problems arised from its freedom and lack of control began to appear gradually,affecting the effectiveness of its' use.In order to achieve regulatory tagging and improve tags' quality,tag recommendations have received more and more attention.Under this background,based on an in-depth analysis of the current tag recommendation systems,including the analysis of its implementation principles and advantages and disadvantages,this article aims at how to integrate multiple tag's sources to improve the tag recommendation results more comprehensiveness and accuracy,taking the hybrid recommendation method as the breakthrough point,combining the current mainstream personalized recommendation technology,introducing the LDA topic model and trust mechanism at the same time,constructing a multi-source tag recommendation model,and integrating the three recommendation methods that based on the recommendation of resource content,collaborative filtering based on resources,and collaborative filtering based on users together form the final recommendation result,and elaborating the implementation method of the recommendation model.Finally,through "Douban Reading" to complete relevant empirical studies.This article is divided into six chapters:The first chapter expatiates on the research situation at home and abroad,then clarifies the research contents and methods of this paper.The second chapter introduces the theory of social tagging system,current personalized recommendation technology and trust mechanism,including the definition,application,recommendation principle of the social tagging system,the definition and nature of several traditional personalized recommendation technologies and trust mechanisms.The third chapter discusses the related content,conceptual basis and working principle of several topic models.It also introduces some current similarity calculation methods based on similarity calculation based on LDA,and paves the way for the construction of subsequent models.The fourth chapter builds a tag hyThere are three recommended methods:the first is content-based tag recommendation,brid recommendation model and elaborates on it.which uses TF-IDF to extract keywords from the content of this resource as a candidate for tagging;the second is based on similar resources' tag recommendations,mainly using LDA topic model to train the text content of the resource to calculate the similarity of each resource,and a tag candidate list based on the resource is obtained.;the third is based on similar user's tag recommendation,mainly using LDA to train the users on topic model,thereby obtain the users' similarity,and introduce the trust mechanism at the same time,then obtain the user-based tag candidate list based on the user's degree of similarity and the degree of confidence.Finally,the three candidate lists are combined to obtain the final tag recommendation results.The fifth chapter is the empirical stage of the model.Taking the "Douban Reading"as an object,crawl the data from the web,then combine with the steps of the model presented in this paper to conduct an empirical study to verify the validity and usability of the model.The sixth chapter summarizes the full text and puts forward the prospect of the next research after analyzing the shortcomings of this paper.