Integrated Mining of Social and Collaborative Information for Music Recommendation

The rapid growth of information enables a large increase in needs of recommendation. To satisfy this need, Collaborative Filtering and Social Filtering are adopted as solutions for personalized music recommendation. Contemporary recommender systems based on collaborative filtering and social filtering always suffer problems of unreliable item similarity, rating sparsity, lack of rating and un-robust evaluation measure. For problems of un-reliable item similarity, rating sparsity and lack of rating, in this paper, we propose an innovative recommender system named Music Recommendation by Social and Collaborative Filtering (MRSCF) by considering playcount and tag information simultaneously. For problem of un-robust evaluation measure, we employ Recommendation evaluation by Normalized Discount Cumulative Gain (RNDCG) to make the evaluation more solid. The experimental results show that, our proposed approach can achieve the higher quality of music recommendation than other state-of-the-art recommender systems in terms of RNDCG and RMSE (Root Mean Square Error).