On the Use of Bipolar Scales in Preference-Based Recommender Systems

Recommendations in e–commerce collaborative filtering are based on predicting the preference of a user for a given item according to historical records of other user’s preferences. This entails that the interpretation of user ratings are embodied in the prediction of preferences, so that such interpretation should be carefully studied. In this paper, the use of bipolar scales and aggregation procedures are experimentally compared to their unipolar counterparts, evaluating the adequacy of both techniques with regards to the human interpretation of rating scales. Results point out that bipolarity is closer to the human interpretation of opinions, which impacts the selection of recommended items.