Pairwise Comparisons in a Logic-Based Recommender System

In this paper, we propose a recommender system using pair- wise comparisons as the main source of information in the user pref- erence elicitation process. We use a logic-based approach implemented in APARELL, an inductive learner modelling the user's preferences in description logic. A within-subject preliminary user study with a large dataset from a real-world domain (car retail) was conducted to compare pairwise comparison interface to one using standard product list search. The results show the users' preference for the interface based on pairwise comparisons, which has proven signifcantly better in a number of ways.