Confabulation based recomender system

A recommender system helps users to make choices among alternative candidates based on their previous behaviors. These systems employ data mining tasks to analyze user data and extract useful information for the prediction of user preferences. Collaborative filtering based approaches recommend to each user according to his/her most similar users. A similarity measure used in these recommender systems affects their performance. In this paper, a new recommender system is proposed that does not need the similarity measure. It recommends using confabulation theory, the user´s previous rating values and other users´ rating values. We compare our system with a collaborative filtering based recommender system which uses Pearson similarity measure. Our experiments on MovieLense dataset show the superiority of the proposed algorithm in term of MAE.