A novel framework for augmenting the quality of explanations in recommender systems

Article type: Research Article

Authors: Karacapilidis, Nikos^* | Malefaki, Sonia | Charissiadis, Andreas

Affiliations: IMIS Lab, MEAD, University of Patras, 26504 Rio Patras, Greece

Correspondence: [*] Corresponding author: Nikos Karacapilidis, IMIS Lab, MEAD, University of Patras, 26504 Rio Patras, Greece. E-mail:karacap@upatras.gr

Abstract: A significant challenge being faced in recommender systems research concerns the provision of robust explanations about why a particular option is suggested. These explanations may exploit diverse data types concerning the users and items under consideration. In line with the above, this paper introduces a novel framework for automatic explanations building in recommender systems. The proposed solution follows a hybrid approach that meaningfully integrates collaborative filtering and sentiment analysis features into classical multi-attribute based ranking. A comprehensive evaluation of the proposed solution advocates the exploitation of additional and diverse information in explanation building, since this better fulfils a series of recommendation related aims such as transparency, persuasiveness, effectiveness and satisfaction.

Keywords: Recommender systems, explanations, collaborative filtering, sentiment analysis, multi-attribute decision making

DOI: 10.3233/IDT-170287

Journal: Intelligent Decision Technologies, vol. 11, no. 2, pp. 187-197, 2017