A Comparison on similarity measures for Rating and Trust based Algorithms for Recommender Systems
In an era of Information age, recommender system helps users to make effective decision. Collaborative filtering is one of the techniques to provide personalized recommendation to users. CF based recommender technique provide recommendation by aggregating ratings from similar users. However, there are a lot of issues in CF for e.g. data sparsity and cold start, which can be removed by incorporating trust information.
CF based recommender technique provide recommendation by aggregating ratings from similar users. To calculate user similarity various methods have been proposed in literature survey e.g. Cosine Similarity (COS), PCC (Pearson correlation coefficient) both of them consider only direction of rating vectors. While in probabilistic similarity measure both the length as well as direction is taken into consideration. In this paper we propose a framework to compare similarity measures in the context of rating-trust based recommender algorithm TrustMF with only rating based recommender algorithm SVD++ and find that rating based algorithm SVD++ outperform TrustMF. To compare similarity measures on these two algorithms we propose a framework and evaluate the result on Film Trust data set.