A Pareto Dominance Approach to Multi-Criteria Recommender System Using Genetic Algorithm
Abstract - Recommender systems are software tools which are used for dealing with the information overload problem by identifying more relevant items to users based on their past preferences. Single Collaborative Filtering, the most successful recommendation technique, provides appropriate suggestions to users based on their similar neighbors through the utilization of overall ratings given by users. But it can select less representative users as neighbors of the active user, indicating that the recommendations thus made are not sufficiently precise in the context of single-criteria Collaborative Filtering. Incorporating multi-criteria ratings into Collaborative Filtering presents an opportunity to improve the recommendation quality because it represents the user preferences more efficiently. However, learning optimal weights to different criteria for users is a major concern in designing multi-criteria recommendation framework. Our work in this paper is an attempt towards introducing multi-criteria recommendation strategies exploring both the concepts of Pareto dominance and Genetic algorithm, to further enhance their quality of recommendations. The contributions of this paper are two fold: First, we develop a Multi-criteria Recommender system using Pareto-dominance methodology (MCRS-PDM). The use of Pareto dominance in our method is to filter out less representative users before the neighborhood selection process while retaining the most promising ones. Second, we applied Genetic algorithm to our proposed methodology for efficiently learning the weights of various criteria of an item. Effectiveness of our proposed RSs is demonstrated through experimental results in terms of various performance measures using Yahoo! movies dataset.
Keywords - Recommender Systems, Collaborative Filtering; Multi-criteria Decision making; Pareto dominance; Genetic Algorithm