Recommendation algorithm based on latent semantic of tags

Abstract

Aiming at the problem that traditional based on user ratings collaborative filtering recommendation algorithm only considering the user’s history rating information, does not Preprocessing user-item rating matrix, leads to the sparse serious of the rating matrix. To solve this problem, puts forward an based on latent semantic tags recommendation algorithm. Firstly, the label attributes of user’s directly or indirectly are used to calculate the similarity between users. Secondly, use the user’s first M nearest neighbors to predicted ratings of items which were not rated and user’s predicted ratings is filled into the user-item rating matrix. Finally, combined with collaborative filtering algorithm to make the top-N interested items are recommended to the target user. The experimental results on MovieLens dataset show that compared with the traditional recommendation algorithm based on user ratings, our algorithm not only solves the sparse problem of the rating matrix effectively, but also improves the precision of recommendation.