Research On Collaborative Filtering Recommendation Integrated With Online User Reviews

With the rapid development of Internet technology,the data volume related to commodities on the Internet increases rapidly and the development of commodities also becomes diversified,various and miscellaneous.Especially in the era of big data,because of the information overload problem,it is hard for people to find those things quickly and correctly which they might be interested in.Under this circumstance,as a good method to solve the information overload problem,personalized recommendation technology provides the right services for different users in the right scenario.Personalized recommendation plays an important role in E-commerce,mobile application and Internet advertisement.Among these many personalized recommendation technology,the most widely used and most successful one is the collaborative filtering recommendation technologies,however,the CF algorithm faces the problems of sparsity of the rating data and it only pays attention to users' score data,ignoring the high quality data of users' comments of commodities.Therefore,a fusing method for the traditional collaborative filtering algorithm is put forward;it applies LDA topic model and Rocchio algorithm on the users' comments and takes consideration of the significant topic of LDA to model the users' preference and then we put forward two fusion strategies: similar fusion strategy with rate fusion strategy and two weighting strategies: static weighting strategy with dynamic weighting strategy,to fuse the text information provided by users' reviews into CF algorithm to improve the effectiveness of calculation based on nearest neighbor and distinguish more accurate and effective neighbor users which will improve the quality of the recommendation model.Finally,based on the improved algorithm,by using some public data sets and appropriate evaluation indicators,we conduct some contrast experiment.The results show that fusing algorithm can improve the accuracy of CF algorithm significantly,similar fusion strategy performs better than rate fusion strategy,dynamic weighting strategy gets better result than static weighting strategy and the result can get better by extracting the significant topic.