Cosine based latent factor model for ranking the recommendation

Abstract

The purpose of this paper is to propose a novel latent factor model that generates a ranked list of items in the recommendation list based on prior interaction with system on e-commerce platforms. The ranking of items in recommendation list is exhibited as an optimization model that optimizes the ranking metrics. The latent features of user and items are learnt using cosine based latent factor model which in turn are used to learn the ranking metric. This paper proposes cosine based latent factor model to learn the implicit features, and corresponding surrogate ranking loss function is optimized. Comprehensive evaluation on three benchmark datasets shows the considerable improvement of the proposed model on ranking metric.