Collaborative Filtering For Long Tail Recommendation

The development of Internet and technology continuously reduces the cost of production,resulting in the rapid expansion of goods space and the increasing number of long-tail goods.The increasingly rich maternal world makes people have higher and higher requirements for individuation,instead of mass consumption of hot items.However,in the sea of information,it is difficult for people to find the long tail items they are interested in,and it is also difficult for the long tail market to locate the approprnate user groups.In order to solve the problem of information overload and asymmetry,the recommendation system emerges at the right moment.However,the current common recommendation algorithms tend to recommend popular products for users.Therefore,it is of great practical significance to construct the personalized recommendation model of long-tail items.This paper studies the MovieLens data set.For the sparsity of the user-item matrnx data and the skewness of the item popularity distribution,the Information Theoretic Co-Clustering and K-means Clustering were used.The empirical study shows that clustering algorithm can effectively improve the density of user-item matrnx and the popularnty of long tail items.As a pretreatment step of item-based collaborative filtering algorithm,the clustering algorithm can also significantly improve the effect of long-tail recoImmendation.In addition,Facing the problem of lacking definite positive and negative feedback information in the implicit user-item matrix and the need of recommending long tail items,this paper explores and improves the latent factor model:For each user-item pair in the user-item matrix,guess whether the user is interested in the item according to whether there is a history record of interaction,and give the confidence of guessing according to the popularity of the item.Such improvement provides positive and negative feedback data for the model more reasonably,and punishes the model for it is inclined to extract the characteristics of hot items in the training period,so as to improve the attention to long-tail items.The simulation study shows that the improved latent factor model can significantly improve the recommendation effect of long tail items while maintaining a high recommendation accuracy rate.As a result,more long tail items are recommended to at least one user,which achieves the purpose of this paper well.