A Hybrid Recommender System In Multi-Criteria Environment

The past decades have witnessed the explosion of information. While we enjoy theconvenience of full information sharing, information overload has been widelyrecognized and experienced. Recommender systems are a specific type of informationfiltering technique and an important method to overcome this problem by providingpersonalized recommendations.Previous studies have pointed out the importance of improving the methods ofuser/item feature representation which still poses challenges to users’ preferencesmodeling. This paper will mainly focus on two interesting questions: first, using precisenumbers or linguistic terms to describe users’ preference information; and second, usingsingle or multiple attributes to describe item information.First, we cite and modify an example to illustrate the importance of incorporatingmulti-criteria ratings for recommender systems. Then we further extend the example toexplain how to collect users’ preference information by fuzzy logic.Second, we present the methodology backgrounds of our algorithm. First, weintroduce two recommendation techniques, collaborative filtering algorithm andcontent-based recommender algorithm. Then, we introduce the basic knowledge oflinguistic variables and fuzzy numbers. Next, we explain the reason of acquiring userpreference information with explicit ratings. Last, we introduce three Multiple CriteriaDecision Making (MCDM) methods--Decision Making Trial and EvaluationLaboratory (DEMATEL), Analytic Hierarchy Process (AHP) and Technique for OrderPreference by Similarity to an Ideal Solution (TOPSIS) which are employed forrecommendation in this research.Then, we propose a new hybrid recommendation approach based on Fuzzy MultipleCriteria Decision Making (FMCDM) methods. Three MCDM methods, DEMATEL,AHP and TOPSIS are integrated with collaborative filtering and content-basedrecommendations.Last, we set up an on line research to collect real life data set. Active users ofDouban.com are invited to rate movies in terms of their story, direction, action and visuals. We ended up with a multi-criteria rating dataset that includes335users and200movies. MAE, TOP-K Hit Rate and scales measure users’ satisfaction in Marketinghave been used to evaluate the performance of the proposed method. The experimentalresults show significant improvement in recommendation accuracy and users’satisfaction.