A Recommender System Based On User Influence And Hidden Factors

As one of the most popular recommendation approaches,traditional collaborative filtering will encounter several difficulties given the problem of data sparsity.Most existing systems mainly consider auxiliary information to alleviate this problem,but they do not make full use of each type of information.To better solve the data sparsity problem and the shortcomings of the existing approaches,we propose an algorithm based on user influence and hidden factors.First of all,we use the PageRank algorithm to calculate the global influence of the user.We then repeat the reviews according to the influence.The intuition behind our repeated reviews is that features discussed by a influential user are more likely to be recommendable features.The greater the user influence value is,the more important roles the related reviews will play in topic model.In other words,we can spread the opinion of influential users by repeating reviews r times.In addition we exploit repeated reviews,as well as ratings,to model user preferences and item features in a shared topic space and subsequently introduce them into a matrix factorization model for recommendation.Experimental results on epinions dataset demonstrate that our approach significantly benefits much from boosted reviews on top-N recommendation tasks compared with various state-of-the-art models,such as Latent Factor Model(LFM),Hidden Factors and Hidden Topics(HFT),Collaborative Topic Modeling(CTR)and Rating-Boosted Latent Topics(RBLT).