Research On Collaborative Filtering Recommendation Algorithm Based On User Clustering

Recently, with the continuous development of e-business and the promotion of the technology of "internet+" as well as the application of the big data and mobile internet, the information has increased in form of explosion within a short term. Under certain situations, it can not satisfy users' demand. Therefore, it has become the urgent problem how people quickly dig the information which is useful to them from the large quantitative one. Thus, at this background, recommendation system application falls into the mainstream, and it can guide people to select all kinds of items they need from the sea of information such as goods, information, etc. Due to its several advantages, the recommendation system has been successfully applied into all sorts of fields, especially e-business, to which the system has brought huge commercial value. As a consequence, its development has also been paid much attention from all circles. However, it indeed has some limits, for example, cold start, data sparsity, and scalability. Targeting these issues, scholars proceed a series of studies. Starting from these problems, the paper will explore the data sparsity, and scalability.This thesis starts from the Recommender System, on the base of analying the temporary researching situation, concretely illustrates the traditional Recommender System and its related theoritical foundation.In addition, the paper also emphases the analysis and introduction of the Collaborstive Filteration. The author points out weaknesses exsiting in the Collaborstive Filteration Recommender System, and improving methods proposed by scholars, when trying to command the knowlege of Recommender Calculation. Due to these, the author also shows readers the design of the thesis: as for the Scalability problems, the author represents the Collaborative Filteration based on clustering users, which means that on the foundation of the users' characterisrics, cluster offline users to solve the problems of the traditional Recommender Calculation; according to the Sparity, the author combination of the existing research results,proposes calculation formula of user similarity based on their rating similarity and project preference similarity calculated by hsrmonic factor.Finally, through the relative comparison with the traditional one, the paper concludes that the Recommender Calculation rooting in user clustering and project similarity has obvious advantages over the traditional one.