Study on data sparsity in social network-based recommender system

Abstract

With the development of information technology and the expanding of information resources, it is more difficult for people to get the information that they are really interested in, which is so-called information overload. Recommender systems are regarded as an important approach to deal with information overload, because it can predict users' preferences according to users' records. Matrix factorisation is very successful in recommender systems, but it faces the terrible problem of data sparsity. In this paper, the authors deal with the sparsity problem from the perspective of adding more kinds of information from social networks, such as friendships and tags into the recommending model in order to alleviate the sparsity problem. The paper also validates the impacts of users' friendships, tags and neighbours of items on reducing the sparseness of the data and improving the accuracy of recommending by the experiments using the dataset from real life.