The Research On Recommender System Based On Topic Recommendation In Collaborative Tagging System

With the rapid development of the Internet and Web2.0 technology, the collaborative tagging systems are becoming increasing popular, and the information overloading is still a severe problem. The research of the recommender system has draw attention for just a short time, the traditional recommendation approaches cannot resolve the information overloading problem properly, we considers the characteristic of the collaborative tagging systems in this paper and propose a new strategy of the recommendation method to fit the collaborative tagging system.The existing recommendation algorithms based on tags are choose the feature of the resources to represent the interests of users, the user profile is always modeled by the vector of the tags, but it cannot describe the actual interest of the users. In this paper, we analysis the users' patterns in collaborative tagging system, including how a user choose tags to describe resources; the pattern of how users' interests change; the relation between tags etc. Considering the conclusions of the analysis, a new method of modeling users' profile using topic which is a collection of tags to represent a user's interest has been proposed in this paper, on the basis the new user profile we modeled and considering the users' interests change pattern, we design a new recommendation algorithm using time window to split the users' history record, in this way, the latest interests using the specific tags of users can be retrieved. The final recommendation can be retrieve in the collaborative tagging system using the tags.In the experiment, we demonstrate that our algorithm is outperforming the algorithm without using time window. Besides, we also discuss the advantages and disadvantages of the method using in this paper.