CC-MOEA: A Parallel Multi-objective Evolutionary Algorithm for Recommendation Systems

Abstract

Recommender systems have nowadays been widely used in a variety of applications such as Amazon and Ebay. Traditional recommendation techniques mainly focus on recommendation accuracy only. In reality, other metrics such as diversity and novelty also play a key role for modern recommendation systems. Although some works based on multi-objective evolutionary algorithm have been proposed for multi-objective recommendation, they are usually very time-consuming because of the large data size of the RSs and the long-term evolution iterations and hence it greatly limits their application in practice. To address these shortcomings, this paper first designs a multi-objective recommendation system, taking into account diversity and novelty as well as accuracy. Then, a novel parallel multi-objective evolutionary algorithm called CC-MOEA is proposed to optimize these conflicting metrics. CC-MOEA is devised grounded on NSGA-II and a cooperative coevolutionary island model, and a parallel global non-dominated selection method is introduced to reduce the runtime of finding the global optimal individuals. Furthermore, a new initialization method and a crossover operator are specifically designed. The experimental results reveal that CC-MOEA outperforms some state-of-the-art algorithms in terms of hypervolume and runtime.