Research And Implementation Of Recommender System Based On Improved Collaborative Filtering

Recommender system is an important means to solve the "information explosion" problem and enhance user experience.On the one hand,it can help users find items they need conveniently and quickly,on the other hand,it can help merchant expend sales volume.Collaborative filtering(CF)is a typical approach of constructing the recommender system.At present,CF still has not ideal effect on sparse data.In this thesis,based on the further study of CF algorithm,we have an exploration for its existing problems.The main contributions are as following:(1)Explore the relevant theoriesThis thesis analyzes the key problems in the traditional CF algorithm on the basis of the research results of recommendation system,CF and matrix factorization(MF)at home and abroad,and summarizes the solutions of the relevant scholars.(2)Propose improved algorithmAn improved CF algorithm based on matrix factorization is proposed to solve the problem of poor performance on sparse data.First,the original rating matrix is filled,and we utilize singular value decomposition(SVD)approach to predict the missing values,then the similarity calculation method combined inverse frequency(inverted file frequency)is used to calculate the similarity between users.Experimental results verify the effectiveness of the algorithm.(3)Build the prototype systemBased on analyzing the key technologies of JavaWeb,combining with the three frameworks and the improved algorithm of this thesis,we build the prototype system of film recommendation.The system test verifies the stability and robustness of the prototype system,which is beneficial to the development of the recommended system in the industry.