Research On Personalized Recommendation Algorithm Based On User Clustering

Enter the 21 st century,the Internet technology improved rapidly,especially the popularity of mobile Internet based on mobile client to allow people to more easily and quickly access to a variety of information on the network,People are gradually moving from the information-deficient era to the information-overload era.Information overload makes the information rich on the network but it makes internet received great challenges,information consumers cannot quickly find the information they are interested in,and information producers are eager to produce their own information by the user's favor.Personalized recommendation system has become the most ideal tool to solve this contradiction,it through the analysis of the user's historical behavior information,learning the user's interest characteristics and thus take the initiative to recommend.However,the personalized recommendation system still faces a variety of problems.This thesis describes the background and significance of the research,summarizes the status quo of the development of the recommended system,and analyzes the advantages and disadvantages of the existing methods of the personalized recommendation system.In view of the shortcoming of the traditional algorithms,such as lack of diversity,an improved algorithm is proposed.This algorithm effectively improved the recommended diversity,and better meets the user's personalized needs and good experience.In accordance with the logic of the thesis,he main work of this paper is summarized as the following three points:(1)Analyze and summarize the related theory and technology of current recommendation system application,and introduce the relevant data set,experimental method and evaluation indicators in detail.In-depth analysis of the traditional recommended method of the principle and the shortcomings of the recommended diversity,and targeted raised solution methods.(2)In view of the shortcomings of the traditional recommendation method of recommendation diversity and lack of novelty,a new algorithm is proposed.For this eigenvector,the algorithm first clusters users aggressively,and according to the degree of belonging,users may be clustered into multiple clusters,which effectively increases the recommended diversity.At the same time the algorithm reduces the contribution of active users and popular objects to the user similarity calculation.Finally,the algorithm effectively improved the recommended diversity.(3)According to the characteristics of the algorithm,a series of experiments were conducted based on the published data set,the result show that this method is feasible and effective.Finally,designed a small recommendation system.