User Preference Representation Based on Psychometric Models

Neighbourhood-based collaborative filtering is one of the most popular recommendation techniques, and has been applied successfully in various fields. User ratings are often used by neighbourhood-based collaborative filtering to compute the similarity between two users or items, but, user ratings may not always be representatives of their true preferences, resulting in unreliable similarity information and poor recommendation. To solve these problems, this paper proposes to use latent preferences for neighbourhood-based collaborative filtering instead of user ratings. Latent preferences are based on user latent interest estimated from ratings through a psychometric model. Experimental results show that latent preferences can improve the recommendation accuracy and coverage while lessening the prediction time of neighbourhood-based collaborative filtering by finding out reliable and effective neighbours; and latent preferences are better than user ratings for representing user preferences.