Research On User-Based Collaborative Filtering Recommendation Algorithm

The technology of personalized recommendation is aimed at studying the behaviors of users, analyzing what they may be interested in and recommending suitable resources to users. In other words, the personalized recommendation is to better solve the contradiction between the requirements of users and the explosive information on the Internet. Collaborative Filtering algorithm based on neighborhood approach and latent factors models is the most successful technology for recommendation system. Although the former is easy to apply, its accuracy needs to be improved. Meanwhile, the latter has high precision, but the model is complex and the parameters are difficult to learn. Therefore an improved collaborative filtering algorithm based on user similarity is proposed. Through adjusting the measure method of user similarity, it can generate more reasonable user neighbors and assure users to evaluate and recommended resources. Experimental results show that the algorithm proposed in this paper is easier to do than the method based on the latent factors models. At the same time, compared to the method based on the neighborhood approach, the algorithm, to some extent, improve the accuracy.Data sparseness is the quality bottleneck of recommender systems. With the increasing number of users and projects, sparseness problem will become more and more serious. Matrix filling method is a common way to solve the problem of data sparseness, while the effect of simple filling method is always not very good. Based on the similarity of different projects, this paper puts forward a Matrix filling method, which fills the sparse matrix with user’s prediction score on projects by recursive computation. Experimental results show that Matrix filling method can effectively relieve the data sparseness problem in recommendation system.