Improving The Recommender Algorithms With The Detected Communities In Bipartite Networks

As the information technology advancing development, human beings are entering to network-oriented information age. The explosive increasing of the data on Internet not only can make our life more convenient, but also brings a serious information overload problem: too much data to effectively extract the valuable ones that reduces the efficiency in making use of information. Recommender system offers a powerful tool to make information overload problem well solved and thus gains wide concerns of scholars and engineers. A key challenge is how to make recommendations more accurate and personalized.In this paper, we notice that community structures widely exist in many real networks, which could significantly affect the recommendation results. By incorporating the information of detected communities in the recommendation algorithms, an improved recommendation approach for the networks with communities is proposed. The approach is examined in both artificial and real networks, the results show that the improvement on accuracy and diversity can be 20% and 7%, respectively. This reveals that it is beneficial to classify the nodes based on the inherent properties in recommender systems.However, in the research process of recommender system, we not only consider the accuracy, diversity and novelty of the results, but also the stability. In the same time, the results of on-line test and laboratory often cannot be completely matched. Then we discussed the top-n similarity recommender algorithm and the top-n stability recommender algorithm, and the experiments show that the top-n stability algorithm can solve this problem. However, this method considers stability more and accuracy less. In this paper, by considering community structures, we propose a stable personalized recommender algorithm. The results show that the stable personalized recommender algorithm can enhance the accuracy and novelty while keep its stability.