Research On The Modeling And Related Algorithms Of Tag Based Recommender System

There exit a new type of system named Collaborative Tagging System in the web as the application of Web2.0 technology. Delicious, Flickr, Youtube, Facebook are the most popular collaborative tagging systems. Users in the collaborative systems can upload or create items to share or for personal purpose, besides items are arrowed to be annotated by arbitrary tags. Because of the UGC(User Generate Content) and free tagging, collaborative tagging systems are suffered by'Information Overload','garbage tags'and 'low precision of search result'. To overcome these problems, recommendation systems is provided. In this paper, we will focus on tag based recommender system.Traditional collaborative recommendation system is based on user-rating matrix which is focus on a bidimensional system. But collaborative tagging system is a tridimensional systems which contains users, tags, items. So we transfer the tridimensional system to three bidimensional systems through projecting:user-user, item-item, tag-tag. For user-user, the paper provide the modeling of user based on tags, IF* IDF, the algorithm to evaluate the rating of user of item, and a collaborative algorithm based on user's interest partition; For item-item, the algorithm to compute the similarity between items and item-based collaborative algorithm are provided; And for tag-tag, the algorithm to compute the relation between tags, tag recommendation algorithm based on social view and the retrieve algorithm based on the expansion of search condition are discussed.At last, experiments are presented which are used to certify the models and algorithms stated in this paper. User based recommendation strategy is designed according to the result of experiments and analysis.