Analogical proportion-based methods for recommendation – First investigations

Abstract

Analogy making is widely recognized as a powerful kind of common-sense reasoning. This paper primarily addresses the relevance of analogical reasoning for recommender systems, which aim at providing suggestions of interest for end-users. A well-known form of analogy is that of analogical proportions, which are statements of the form “a is to b as c is to d“. Encouraged by good results obtained in classification by analogical proportion-based techniques, we study the potential use of analogy as the main underlying principle for implementing rating prediction algorithms. We investigate two ways of using analogical proportions for that purpose. First, we exploit proportions between four users, each described by their respective ratings on items that they have commonly rated. The second prediction method only relies on pairs of users and pairs of items, and leads to better performances. Finally, in order to know to what extent it may be meaningful to apply analogical methods in data analysis, we address the problem of mining analogical proportions between users (or items) in ratings datasets. Altogether, this paper initiates a general investigation of the potential use of analogical proportions for recommendation purposes.