Research On Hybrid Collaborative Filtering Algorithm Based On K-means User Clustering

In recent years, with the rapid development of electronic commerce and the popularity of the Internet, The lives of people are more and more convenient, and they can select the items they want without leaving home, the e-commerce system also continue to provide consumers with goods and services. However, with the increasing of users, e-commerce system to provide the type of goods information also continue to increase, although the user can easily select a variety of goods, but now it is difficult for users to quickly find accurate goods they are interested in many information, at this point, the improvement and optimization of the electronic commerce recommendation system is imminent.Now a lot of recommendation algorithms already exist, which mainly can be divided into the personalized recommendation based on user’s subjective, and non-personalized recommendation algorithm for user subjective intent irrelevant, and accordingly recommended system can also be divided into non-personalized recommendation and personalization recommended system. Recommended system provides personalized recommendations is to give recommendations based on the historical behavior of the user, rather than a non-personalized recommendation does not consider user history data. This article mainly studies the collaborative filtering in personalized recommendation algorithm, because the traditional collaborative filtering has some disadvantages, this paper do some corresponding improvement against its drawbacks, the main work is as follows:First of all, This paper uses the most traditional and most widely used recommendation algorithm, namely collaborative filtering recommendation algorithm. Collaborative filtering recommendation algorithm is divided into two kinds of collaborative filtering recommendation algorithm based on user and collaborative filtering recommendation algorithm based on item. the first way is recommended by calculating the similarity between users and the other is recommended by calculating the similarity between items, This paper mainly uses the user-based collaborative filtering.Secondly, This paper uses hybrid collaborative filtering recommendation algorithm. This paper is optimized for sparsity of user-based collaborative filtering algorithm. In real life, the user score is only a small part of the item, making the rating data is very sparse, which will lead to big error exists in the calculation of similarity of users, it is very difficult to accurately find the user’s "neighbors", a direct impact on the accuracy of the recommendation results. This paper according to the similarity of the item to predict score, solve the sparse of the rating matrix, then according to the neighbor users’ behavior to recommend.Finally, due to the rapid development of e-commerce, users and items are increasing, leading to calculate the similarity appears low efficiency, it is recommended not timely issues when using collaborative filtering recommendation algorithm, therefore, this paper uses the data mining algorithm. firstly using k-means user clustering, the users are divided into different clusters according to the user’s behavior, then using the project similarity score to solve the sparsity of the matrix, finally on the basis of this, using collaborative filtering recommendation algorithm, we should determine the user belongs to which cluster, and then in the user belongs to cluster computing similarity. so that to maximize the reduced amount of computation, and improve the efficiency of recommendation algorithm, and according to the experimental results, this optimization method will improve the accuracy of the algorithm.