Automatic user preference elicitation for music recommendation

Recommendation Systems involve effort from the user to elicit their preference for the item to be recommended. The contribution of this paper is in eliminating such effort by automatically assessing user's personality and using the personality scores for recommending music tracks to them. Automatic personality assessment is performed by automatically answering a personality questionnaire by observing user's audiovisual recordings. To obtain personality scores, traditionally the answers to the questionnaire are combined using a set of rules specific to the questionnaire to get personality scores. As a second contribution, an approach is proposed to automatically predict personality scores from answers to a questionnaire when the rules to combine the answers may not be known. Promising results on a dataset of 50 movie characters support the proposed approaches.