Research On Collaborative Filtering Algorithm In Personalized Recommendation

With the rapid development of information technology and network technology, people have more ways to get information, but the fact that the explosion of information in the network lead to the users lost in the ocean of information, it is more difficult to select the information they really need, the phenomenon which is called the information overload. In order to solve the problems, personalized recommendation system coming out, it does not need users to input any information initiatively, by building user interests model through the analysis of user behavior history, recommend to the user the information they may interest initiatively. The core of the personalized recommendation system is the recommendation algorithm, collaborative filtering algorithm is the most studied and the most widely used recommendation algorithm. The workflow of collaborative filtering recommendation algorithm will be analyzed in this paper in detail, and an improved collaborative filtering recommendation algorithm is proposed to improve the recommendation quality of the recommendation system for the data sparse problem and the interest migration problem.The main research work of this paper:(1) According to the sparsity of rating data matrix, this paper proposes an improved collaborative filtering algorithm. At first, cluster the entire project set according to the project attributes, and use the slope one algorithm for filling, calculate the similarity of users using the weighted similarity of users in each cluster.(2) The traditional collaborative filtering algorithm relying on rating data to recommend, which does not take into account the factor that the user’s interest with the time to change, the earlier the score, the lower the value, in order to predict the score more accurately, this paper introduces the Ebbinghaus forgetting rule into the recommendation process, by adding a time weight for each score to improve the quality of the recommendation system.(3) In order to prevent that there is only a few neighbors give score of the target item, calculate the score by similarity between the target user and each neighbor at first, and then get a virtual nearest neighbor matrix, use the similarity again on the virtual nearest neighbor matrix to predict the score.At last, in order to verify the improved algorithm is effective, the movielens data set respectively on the traditional collaborative filtering algorithm and the improved algorithm are tested and compared, the experimental results show that the improved algorithm has better effect.