A NOVEL WEIGHTING SCHEME FOR A MULTI-CRITERIA RATING RECOMMENDER SYSTEM

摘要

Recommender systems aim to introduce interesting items to a particular user. Most use a single rating or overall score representing the user's overall preference toward each item to produce a recommendation. Recently, many researchers have proved that multi-criteria ratings provide better results than a single criterion. According to the concept of multicriteria rating recommender systems, the criteria weight is integrated with the user profile to identify the importance the user attaches to each criterion. If a suitable weight is obtained, the user characteristics will be more correctly represented. Consequently, high-quality neighbors and high-accuracy recommendations will be formed. Nowadays, weight assignment is developed automatically and adaptively, but does not consider the detailed rating information about how much all possible rating scores affect the user preference on each criterion. In this work, a new weighting scheme is proposed based on the assumption that considering detailed rating information will produce a better recommendation. The new weighting scheme, called pf-ipf (preference frequency - inverse preference frequency), counts the frequency of possible individual rating score occurrences instead of words in the concept of tf-idf coming form the text-mining. Experimental evaluation shows that our pf-ipf technique is effective.