Research Of Sparse Problem In Collaborative Filtering Based On Tag

The rapid development of Internet and e-commerce has greatly expanded the space for the user to select products. But also brought the problem of information overload. Users have to navigate much irrelevant information before finding the goods they need. In this case, e-commerce recommendation system came into being, it forecasts the user's interests and provides users with personalized product recommendation service.Collaborative filtering is a recommend technology based on user behavior. It typically use a user-item rating matrix to find the current user's nearest neighbors, and then predict the current user's preferences by referring to these nearest neighbors'preference, and recommend a group of products which are most likely to be bought by the current user. Although collaborative filtering has been widely used and have excellent performance in recommendation, it is also facing the challenge of sparsity.In collaborative filtering systems, the rating given by a user to an item is stored in a two-dimensional matrix, but only very few users provide rating to the system. With rapid increase in the number of users and items, a lot of and even the majority of the matrix elements are empty, which formed a sparse matrix. The sparse matrix has a negative impact on the accuracy of recommendation in three aspects——similarity calculation, selection the nearest neighbors and prediction.After detailed analyzing to the sparse problems, we proposed two hybrid recommendation methods TAG-CF and improved TAG-CF which are both based on collaborative filtering and tag. These methods transfer the sparse user-item matrix to a relative dense pseudo matrix by filling the prediction ratings based on tags. Experiment results show that the proposed method performs significantly better than the traditional CF method. Based on the theoretical research, we designed an unified framework for personalized recommendation system and integrated different recommendation approaches into this framework to meet the diversified needs of different users in different scenarios.