Correcting Popularity Bias by Enhancing Recommendation Neutrality.

Abstract

In this paper, we attempt to correct a popularity bias, which is the tendency for popular items to be recommended more frequently, by enhancing recommendation neutrality. Rec-ommendation neutrality involves excluding specified infor-mation from the prediction process of recommendation. This neutrality was formalized as the statistical independence be-tween a recommendation result and the specified informa-tion, and we developed a recommendation algorithm that satisfies this independence constraint. We correct the popu-larity bias by enhancing neutrality with respect to informa-tion regarding whether candidate items are popular or not. We empirically show that a popularity bias in the predicted preference scores can be corrected.