Research On Collaborative Filtering Recommender Algorithms Based On User

With the arrival and rapid development of Web 2.0,the internet information becomes into the explosive growth, which makes the problem of information overload highlight. The massive data makes users difficult to find useful information. At present, search engine and recommendation system are the main means to solve the problem of information overload. Search engines provide keyword-based undifferentiated service, which means people will receive the same results when using the same keywords.The collaborative filtering recommendation algorithm is the the most popular and successful recommendation algorithm currently. The algorithm excavates the user preferences from the historical information, without analyzing the specific content. The recommendation process has three links:the similarity calculation, the neighbor selection and the recommendation according to the neighbors. With the deepening research of the collaborative filtering recommendation algorithm at home and abroad in recent years, the various aspects of the process also existe the following deficiencies.(1) It’s not accurate by the single similarity calculation, the trust relationship between the social network users is introduced to alleviate this problem, but there are the problems of complex path selection and weak trust transfer by many nodes route hop.(2) The neighbors are only selected by similarity, without considering the contribution of neighbor users.(3) K- mean weighted recommendation reduces the needs of individual users, and the determination of k is too subjective, that affect the accuracy of recommendation;(4) The collaborative filtering process doesn’t take in the interference.This paper aims at the above problems of collaborative filtering process, the research content and innovation points of this paper are as follows.1. A one-jump trust model based on project was designed and the new trust mechanism was defined, to solve the problems of complex path selection and weak trust transfer, the algorithm defined the consumer’s trust attribute vector of social and excuted the direct and indirect distance O ne-jump by items. Experiments shows algorithm predicts better and saves about half training time.2. To address the problem that the neighbor selection process does not take into account the contribution of neighbors, this paper thinked over the non common evaluation items, introduces the concept of the contribution factor to make the selection of neighbors reasonable. Experiments show the algorithm can improve the recommendation performance;3. A heuristic clustering model was proposed to determine the neighbor number k subjectively, and the category similarity was introduced to adjust the K weighted mean algorithm. Experimental results show the algorithm improves the recommendation accuracy;4. A low pass filtering recommendation model was designed and the noise and trust in recommender systems ware defined, to determine the similarity calculation weight based on the significance of rating the users to the items, which avoided the problem that small significance items got the big weight in the similarity calculation. Experiments show that the improved algorithm has higher performance.