Research On Collaborative Recommender System Incorporating Social Tags

Recommender System is a useful tool for the solution to information overload. Traditional Recommender System is based on ratings, scarcely paying attention to the social tags. But now, applying social tags to the recommendation has become a tendency. Because of abundant information social tags have, they can be an effective way to express user's interest and product's feature. If social tags can be incorporated with ratings in a Recommender System, many benefits can be obtained including the deep information in data, the improved quality of recommendation, and the satisfaction of personal needs.In order to have a quick search for knowledge and achieve a high quality of recommendation, some primary tasks have been done based on the knowledge of social tags and the personal recommendation techniques, as well as the collaborative filtering algorithm incorporating social tags is taken as a research target.1. The theory and method are studied and summarized that incorporate tag information with ratings. On the base of that, the method of a rating matrix extended by social tags is regarded as the research focus.2. A method is proposed that incorporating latent variables in social tags with ratings. The triple of(user, tag, item) can be broken up into two binary tuples that are(user, tag) and(item, tag) respectively. At the same time, the TF-IDF(Term Frequency-Inverse Document Frequency) is taken as the value of two binary tuples. On the base of that, LFM(Latent Factor Model) is used to extract the latent eigenvector of users and items. Next, two incorporation methods are executed on the latent eigenvector and rating, which are vector enlargement method and linear weighted similarity method. At last, the rating prediction is made with the memory-based collaborative algorithm.3. A comparable experiment between the traditional collaborative recommender algorithm and the proposed method is conducted on the 100 K Movie Lens dataset. It can be proved that the performance of the proposed method is superior to the traditional one.4. In order to solve the management problem of articles and help user to find personal resources quickly, an article recommender system is designed based on the proposed algorithm. In this system, users are allowed to categorize posted resources freely. At the same time, personal recommendation can be provided with the tag and rating information. As a result, every user can have a better reading experience.