Using multi-objective optimization to solve the long tail problem in recommender system

Abstract

An improved algorithm for recommender system is proposed in this paper where not only accuracy but also comprehensiveness of recommendation items is considered. We use a weighted similarity measure based on non-dominated sorting genetic algorithm II (NSGA-II). The solution of optimal weight vector is transformed into the multi-objective optimization problem. Both accuracy and coverage are taken as the objective functions simultaneously. Experimental results show that the proposed algorithm improves the coverage while the accuracy is kept.