Accuracy improvements for multi-criteria recommender systems

Abstract

Recommender systems (RS) have shown to be valuable tools on e-commerce sites which help the customers identify the most relevant items within large product catalogs. In systems that rely on collaborative filtering, the generation of the product recommendations is based on ratings provided by the user community. While in many domains users are only allowed to attach an overall rating to the items, increasingly more online platforms allow their customers to evaluate the available items along different dimensions. Previous work has shown that these criteria ratings contain valuable information that can be exploited in the recommendation process.