Strategies For Collaborative Filtering Recommendation Based On Social Tags

In the Internet world, while people access to the information, they are also providing information to others. Therefore, how to find valuable information from the vast amounts of information in order to meet the user's needs, and how to find and enjoy the valuable information by the required users, have been a hot issue which is concerned by academia and the business. As a result, the Internet industry produced two very important technologies: search and recommendation technologies.Collaborative filtering (CF) is the most widely recommendation technique. It believed that users interested in the same or similar resources with similar preferences. Thus, the collaborative filtering technology can be used to predict individuals with groups. It can fully mine the wisdom of groups to service the individuals.In this thesis, we study two kinds of traditional collaborative filtering recommendation algorithms - User-based Collaborative Filtering and Item-based Collaborative Filtering. And point out the problems existed in the traditional collaborative filtering methods: ratings matrix sparsity, cold start, vulnerability, single interest model and scalability issues.With the development of Web2.0 technology, social tag has been widely used in many personalized sites. Using tags to describe resources is more accurate, and it can reflect the real preferences of the individual users. Therefore, social tag is more suitable for the recommendation technique.In this thesis, tag-based collaborative filtering algorithm is proposed, and introduces the social tags into traditional collaborative filtering methods. This new algorithm can mine the potential preferences of users, and then recommend items in the user's preferences scope. This method can improve the traditional collaborative filtering methods, and can solve the single-interest model problem of traditional methods. And this new approach reduces the size of the rating matrix. It makes the computing more efficient. The experiments based on MovieLens data set shows that the tag-based collaborative filtering method is significantly better than the traditional collaborative filtering methods in recommendation effects.