Research And Implementation Of Collaborative Filtering Recommendation Algorithm Which Faced To Sparsity Data

Recommendation system based on collaborative filtering is a very good program to solve the problem of information overload, and provides personalized services. But it also faces a series of bottleneck problems, and the data sparsity is one of them that have the greatest influence to the efficiency of collaborative filtering.There are a large number of items in recommendation system. But every user in recommendation system just rates a few items. And it leads to the sparsity of rating data. And selecting accurate nearest neighbors of users becomes difficult due to the sparsity of rating data. It leads to the inaccuracy of recommendations.In this paper, we do some in-depth research about the problem of data sparsity in recommendation system. The main works are as follows:1. From the perspective of similarity calculation, this paper analyses the disadvantages of traditional similarity measure. We build the neighbor level between target subject and other subjects according to the number of ratings they have both rated, and improve the similarity calculation method based on the neighbor level. Then, a collaborative filtering algorithm based on improved cosine similarity is been proposed. The experimental results show this algorithm can effectively improve the accuracy of recommendation.2. From the perspective of alleviating the sparsity of rating matrix, we analyze the cause of data sparsity, and propose a collaborative filtering algorithm based on vacant rating filling with items and modal method. In order to reduce the sparsity of rating data, we fill the vacancy of rating data before selecting the nearest neighbor of target user. We fill the vacancy based on the original data set. Firstly, we use improved cosine similarity measure the similarity between items to enhance the veracity of the similarity measurement between items. Secondly, we fill the vacancy data with the similarity of the items and modal method to solve the problem of data sparsity. Thirdly, measure the similarity of users using the same method for measuring item similarity based on the filled data to enhance the accuracy of the similarity measurement between users. Last, use the similarity of users to find the active users’ neighbors. The experimental results show that this method can effectively reduce the inaccuracy of neighbors searching in sparse data to a certain extent, and improve the accuracy of recommendation.3. We analyze the cause of cold start problem in recommended system. And we propose a method to solve the cold start problem in collaborative filtering recommendation based on the mode of rating. This method uses different mode of rating to solve the new user and new item problems. The experimental results show that this method can solve the cold start problem to a certain extent.