Data-driven decision making in critique-based recommenders: from a critique to social media data

In the last decade there have been a large number of proposals in the field of Critique-based Recommenders. Critique-based recommenders are data-driven in their nature since they use a conversational cyclical recommendation process to elicit user feedback. In the literature, the proposals made differ mainly in two aspects: in the source of data and in how this data is analyzed to extract knowledge for providing users with recommendations. In this paper, we propose new algorithms that address these two aspects. Firstly, we propose a new algorithm, called HOR, which integrates several data sources, such as current user preferences (i.e., a critique), product descriptions, previous critiquing sessions by other users, and users’ opinions expressed as ratings on social media web sites. Secondly, we propose adding compatibility and weighting scores to turn user behavior into knowledge to HOR and a previous state-of-the-art approach named HGR to help both algorithms make smarter recommendations. We have evaluated our proposals in two ways: with a simulator and with real users. A comparison of our proposals with state-of-the-art approaches shows that the new recommendation algorithms significantly outperform previous ones.