Researcn Of Personalized Recommendation In Social Tagging Systems Based On The Graph Structure

With the development of Web2.O, Social Tagging System is more and more popular with Internet users. In social tagging systems, users collect interesting resources, annotate and manage them with tags. Since there are a great deal of information in systems, how to find other users with the similar interest, search interesting resources and exploit appropriate tags to manage them is the problem that every user faces. It is necessary for us to research on personalized recommendation in social tagging systems. Effective recommendations are beneficial for improving personalized service of websites as well as helping users get their interesting information.Considering the cold start problem on resources and low precision in tag recommendation, the thesis proposes tag recommendation method based on resource content, which aims to recommend relative tags for the new coming resource in social tagging systems. Firstly, similar resources are selected for the given resource according to the similarity between texts, which is computed based on the graph structure of texts. Secondly, features of the given resource and tags of similar resources are grouped as the candidate tags of the given resource, and then different methods are designed to compute the membership values for the corresponding tags. Finally, the tag extracting method is developed, and tags are recommended to the given resource based on their values. The proposed method can avoid the resource cold start problem. Meanwhile, it expands the recommendation range and introduces new tags to the social tagging system. The experiments investigate the precision, recall and novelty of the social tagging system through different tag extracting methods. The results show the proposed method performs better.Considering tri-relationship among users, resources and tags in social tagging systems, and data sparsity problem in collaborative filtering technique, the thesis proposes a user taste diffusion model based on tripartite graph, aiming to recommend other users with similar preference and other interesting resources to the given user. It makes the given user’s taste diffuse to other users, resources and tags according to multi-diffusion mechanism, and then generates recommendations among other users and resources based on their corresponding user taste values. Experiment results on public data sets show that the proposed user recommendation method outperforms the baseline method in similarity and network density and the proposed resource recommendation method performs best in precision and recall according to comparing with other two methods which are bipartite graph based recommendation method and collaborative filtering based recommendation method.Research of personalized recommendation in social tagging systems not only solves resource cold start problem in tag recommendation and data sparsity problem in collaborative filtering to some extent, but also improves customer satisfaction and lovely.