A recommender system based on trust and semantics in collaborative systems using a new measure of association
Seyedeh Homa Alizadeh, N. R. Arghami, Majid Vafaei Jahan · 2015
One of the most popular techniques used in recommender systems is collaborating filtering. In this technique it is usual that Pearson's correlation is used to find the similarity between users. It is a known fact that Pearson's correlation is not suitable for measuring the strength of nonlinear relations. Since Spearman's correlation is in fact Pearson's correlation applied to ranks and does not work well in non-monotone relationships and since measures like Kendal's tau do not work well in small samples, we introduce a new measure of association to be used in collaborative systems which we shall call alpha. Our investigations show it leads to better MAE. We also propose a method by combining Alpha and trust propagation and add a new algorithm to semantic similarity for confronting the problems with cold startand it leads to better coverage.