Collaborative Filtering TopN-recommendation Algorithm Based On The Improved Similarity Research

With the advent of the Internet, rapid development of networking and cloud computing technology, our life also occurred in the earth shaking changes. We enjoy advanced technology brings convenient at the same time, but also bear the they bring negative influence, information overload is one of them. To solve the problem of information overload early mainly classified directory and search engine two solutions, but has its own limitations, classified catalogue can only cover a small number of projects, with respect to the massive project is far from enough; search engine needs to have a clear demand, unable to meet the user’s individual needs. In order to further solve the problem of information overload, the researchers put forward to promote The concept of recommender system.Recommendation system, through the analysis of user behavior, dig out the hidden user preferences, and to realize the user personalized recommendation. Recommend the core part of the system is the recommended algorithm, now in the actual system use the most widely used algorithm is the collaborative filtering recommendation algorithm. But with the further application of recommender systems, the traditional collaborative filtering recommendation algorithm such as cold start, data sparsity, recommended low precision, long tail project to explore the lower ability of the no longer meet the needs of practical application. At present to recommend system research is very hot, but quite a lot of research work mainly focus on how to improve the High recommendation accuracy, although achieved some results, but can not take into account the novelty of the results of the recommendation, leading to the recommendation of novelty is still in a relatively low level.This paper proposes collaborative filtering TopN recommendation algorithm based on item similarity. The main research purpose of this paper is to enhance the ability of the long tail project to explore and not reduce the accuracy of the results:1. Analysis of the impact of user activity on the recommended results, the introduction of the concept of user activity, weakening the impact of active users on the recommended results, improve the long tail project to explore the ability to improve the coverage of the recommended results.2. Analysis of the types of projects on the recommendation of the impact, the similarity matrix was normalized, weaken the similar project saw the similarity too high on the recommendation of the influence, improve the accuracy of recommendation results, and enhance the reliability of recommendation results.3. The classic MovieLens data set is used to simulate the improved algorithm step by step, and the effectiveness of each step of the algorithm is verified.