A Game-Based Evolutionary Clustering With Historical Information Aggregation for Personal Recommendation

In order to alleviate the network information overload, recommender system becomes widespread in personalized recommendation. However, due to the increase of the number of users and items in the network, the rating data gets increasingly sparse. At this time, it is necessary to use a variety of clustering algorithms to divide nodes into different communities, and then make recommendations in each community, which can improve the performance of recommendations and reduce the time complexity of algorithms. In this paper, we propose a game-based evolutionary clustering with historical information aggregation for personal recommendation. Firstly, a payoff function of game theory is introduced into the evolutionary clustering to accelerate the stability of the algorithm. In this clustering approach, the next state value of each node is related not only to the current state value, but also to the historical states, hence, it achieve better prediction results. Meanwhile, the clustering method is theoretically proved to be stable by Lyapunov stability theory. And then, we predict the possible ratings by the user-based collaborate filtering method, and recommend items for target users according to the preferences of neighbors. Finally, diverse experiments are executed on seven real recommendation datasets to verify our recommendation results are better than several compared algorithms.