Research On Collaborative Filtering Algorithm Based On Item Category And Confidence Matrix Filling

In recent years,with the rapid development of the Internet,the problems brought about by it have become more and more obvious.The most prominent problem is the "information overload" problem.Although technologies such as catalogs and search engines can alleviate the problem to some extent,most users can't clearly know what they need,and it is difficult to use these technologies to search.However,the personalized recommendation not only can discover the user's interest preference based on the user's historical information,but also analyzes the potential interest of the user among the massive data,which attracts much attention.In the development of personalized recommendation system,there are more and more problems,such as data sparsity,cold start,scalability and so on.In this thesis,we first use graph network algorithm to compute the item correlation matrix on sparse data set.Then this thesis introduces a k-dimensional item on the problem that the related item matrix has weak correlation.Each item is a probability distribution on the class vector,and the class vector matrix updates itself by randomly moving around the item association graph,and the updated class vector matrix reflects a stronger item association degree,thus calculating the similarity between items,and then filling in the scoring matrix for the first time.Finally,the concepts of user set classification and scoring confidence are introduced to improve the traditional item-based collaborative filtering algorithm,and the scoring matrix was filled in for the second time.In this thesis,the item-based collaborative filtering algorithm is used as the benchmark algorithm to conduct comparative experiments on the Movielens dataset.The results show that the Collaborative Filtering of Item Category(CFIC)algorithm and the Collaborative Filtering of Item Category and Confidence(CFICC)algorithm has better recommendation effect.