Research On Collaborative Filtering Recommendation Algorithm Based On Optimized Neighborhood

With the development of the internet and information techniques, especially on the rise of the E-commerce, our life begin to accept more and more electronic information, but the rapidly growth of the information resources make people feel confused to choose. As a kind of effective information filtering technology, recommender system using knowledge discovery technology to solve the personalized recommendation problems such as information, goods and services in our life.The collaborative filtering technology can be used to filter the information which is complex concept and difficult to automatically analysis for machine, and the result of recommendation is novelty, so it has been one of the most widely used personalized recommendation technology. However, the traditional collaborative filtering technology face many challenges such as the data sparsity problem, cold start, scalability, real-time, similarity measure and so on, it influences the accuracy of recommendation system.In this paper, because of the data sparsity and the difficult of similarity measure that existing in the traditional collaborative filtering algorithm, we proposed an improvement collaborative filtering algorithm based on optimal neighbor. In the improvement algorithm, firstly, the algorithm according to the similarity of users and items, using dynamic strategies in the neighbor set selection process. The dynamic selection strategy enables user to have different size of the neighbor set based on target user’s similarity. It is effective to avoid the limit of application scenarios and the problem of target user’s difference. This strategy can effectively improve the quality of similar users, and make contribution to improve the accuracy of recommendation.Secondly, in the target user’s neighbor set, we use trust model to calculate the trust value of neighbor users. According to the trust values, we build the optimal neighbor set based on the dynamic selection strategy. The method of trust value calculating is accuracy of rating prediction to the target user’ rating items by neighbor user, thus we can reduce a negative effect on the recommendation of the target user. Trust model is helpful to acquire the user who has ability to recommend to predict score. Finally, in the process of rating prediction, the trust value will instead of similarity values. Based on the sum of trust value in neighborhood set of target user and target item, we determine the percentage of user neighborhood in the prediction process. It can effectively improve the recommendation accuracy and ease the problem of sparse matrix. Experimental results show that, the proposed algorithm can effectively improve the accuracy of recommendation collaborative filtering algorithm.