Research On Collaborative Filtering Algorithm For Rating Prediction With Tag Information

With the rapid development of science and technology,how to obtain useful information from the complex information world and to solve the problem of information overload has become a challenging task.Recommend system,as an effective way to solve the problem of information overload,has become a hot issue in the current academic community.The recommendation algorithm is the core component of the recommendation system.Among them,the collaborative filtering is widely used because it is simple and easy to implement.For the problem of data sparsity and cold start in the recommendation system,effective collaborative filtering algorithms based on neighborhood and matrix decomposition are proposed.When using the neighbor-based collaborative filtering algorithm for rating prediction,the choice of similarity measurement strategies between users and items is the key to the recommendation quality,but the similarity cannot be properly measured by only rating matrix.For this purpose,how to use extra information to improve the similarity between users and projects has become the main task of this article.This thesis focuses on the similarity measurement methods between users and items and the usage of tag and time information.The main work of the thesis is as follows:1.An improved collaborative filtering algorithm unified rating and the item correlation is proposed(URIC).The algorithm integrates the item's attribute relevance and interest relevance into the similarity calculation process to make up for the lack of similarity calculation based only on the rating matrix.Experiments on MovieLens dataset show that the URIC algorithm has good recommendation results and can well alleviate the data sparsity problem.2.An improved collaborative filtering algorithm based on tag information and time factor is proposed(TT-CF).The algorithm combines the tag information and time information to improve the traditional collaborative filtering.The use of tag information and rating information to calculate the similarity between users and items overcomes the deficiency that based only on the rating matrix.It also introduces time weights to give the items which is more recently rated by the target user a larger weight in order to emphasize the user's recent interest behavior.Through comparison experiments conducted on the MovieLens dataset,the results show that the TT-CF algorithm can improve the accuracy of the recommendation.3.An improved collaborative filtering algorithm unifyed the Bhattacharyya coefficient and LDA topic model is proposed(UBL-CF).The algorithm combines the LDA topic model and the Bhattacharyya coefficient to improve the performance of traditional collaborative filtering algorithm.It utilizes the LDA topic modeling method to mine potential topic information on the users'(items')tag set.It introduces the Bhattacharyya coefficient to use all the rating information between users to get rid of the common rating restrictions and handle data sparsity problem.