Models Of User-Based Similarity Collaborative Filtering Recommendation

The rapid development of information technology a nd the current growth and popularity of the Internet have exacerbated the information overload problem. Recommender system is a kind of information filtering system that gives advice on product information or services which a user may be interested in. It adopts knowledge discover techniques to discover user interests according to user behavior and then to make recommendations. The core of the collaborative filtering in a recommender system is the choice of neighbors to the active user. Based on this, we propose two algorithms of user similarity for constructing the similarity model. The main works are as follows:It is reasonable that the sharp boundary problem can be addressed by using fuzzy logic. The fuzzy set concept provides a smooth transition between members and nonmembers of a set. The item neighbor information, which means information such as the rating number and rating values obtained from all the users except the active user in the recommendation system, can be used for building the similarity mode l. Based on these, similarity measure using fuzzy rules and contextual information for collaborative filtering is proposed. Experiments show its superiority on accuracy and recall compared with traditional clustering algorithm.The core of collaborative filtering is to find the neighbors based on the similarity measure. More specially, it uses the most selected similar neighbors as a reference to show recommendation for users. Therefore, a reasonable assumption is that in the most selected neighbor users or items which have a negative impact on the prediction should be r emoved. Additionally, the proposed method should be easily combined with o ther similarity measures. Based on these, a new reasonable similarity function to find the suitable neighborhood of the users and eliminate the noisy neighbors to get a more accurate prediction is proposed. Experiments show that its superiority on accuracy compared with traditional collaborative filtering algorithm.Particle swarm optimization is a population based stochastic optimization technique which has a wide range o f application in many areas. The development of PSO is based on the metaphor of social interaction and communication such as birds flocking and fish schooling. The algorithm is simple, high precision and fast convergent. Based on these, we use the particle swarm optimization to find the weighs of each distance similarity factor. Experiments show its can get better results compared with traditional collaborative filtering algorithm.