Research Of Hybrid Recommendation Algorithm

With the rapid development of information technology and the popularity of the Internet, the network has become the main way to obtain information in people’s daily life. The Internet provides users with more and more information and services. The information resources on the Internet are beginning to show explosive growth trend. When facing massive network resources, users are difficult to find useful information quickly. In this context, personalized recommendation algorithms come into being. Personalized recommendation algorithm can adjust the information services according to users’ personal interests. However, the development of the recommendation algorithm is still not mature and sound in the current. At this moment, there are still many problems in the recommendation algorithms. For example, low accuracy, serious data sparity, cold start and real-time issues.In this paper, to improve the accuracy of recommendation algorithms, we focus on user-based collaborative filtering algorithm and attribute-based algorithm. We compared their advantages and disadvantages between the two algorithms. Then we combined user-based collaborative filtering with the attribute-based recommendation algorithm. The main steps in the hybrid recommendation algorithm are the calculation of similarity, the selection method of nearest neighbors and score prediction method. In order to improve the accuracy of recommendation algorithm we focus on the first two steps. We calculated the similarity based on the approaches proposed by others. There are two main selection methods of nearest neighbors:top-n method and threshold method. Based on these two methods, we proposed a hybrid nearest neighbor selection method. By adjusting the selection of nearest neighbors, we can improve the accuracy of recommendation algorithm. Finally, we do experiment on MovieLens data set and BookCrossing data set. And we compared the proposed method with the original methods. The MAE and RMSE results show us that the proposed methods improved the accuracy of the recommendation algorithm.