A new recommendation technique for interval scaled multi-criteria rating systems incorporating intensity of preferences

Abstract

We present Interval-Rec, a recommender system that gives predictions on items that are rated on multiple criteria. Although a five-star rating system or similar linguistic scales are used typically by on-line sites to enable their users to rate items such as content or products, ratings are considered usually as ordinal and treated as ratio during the calculation of predicted ratings. We demonstrate that these symbolic or lexical semantics convey information about the strength of user preferences in addition to the order of the rated items. The methodology we propose considers and treats such scales as interval and in the same time provide accurate recommendations to users. Evaluations using well-known and reliable data showed improved results over other significant multi-criteria recommender systems and state of the art single criterion method.