Research On Recommendation Method Based On Improved K-means Clustering

Development of the Internet and E-commerce generated a lot of data, resulting in information overload. When the search engine technology has been unable to solve the problem of information overload, the recommender system appears. The common recommender systems are collaborative filtering, Item-based and Graph-based, but they all have sparse data and computationally intensive problems, which cause recommended inaccurate results and poor real-time.In-depth study of these issues, the main work and contributions are as follows:Firstly, randomly chosen initial cluster centers can easily cause the problem of inaccurate clustering results. We propose an alternative method of K-means initial clustering center improvement based on the idea of minimum variance and maximizing the minimum distance and a new method to measure the similarity between users based on user trust relationship and preference ratings, solving the data sparseness problem.Secondly, we propose a recommended method IKC(Improved K-means Clustering Recommendation Method) based on improved K-means clustering, the trust-based similarity instead of K-means Euclidean distance between objects. We use the improved K-means algorithm to cluster users, generating a recommendation list with Top-N in the same sub-class.Finally, experiments on MovieLens(1M) and Epinions datasets show: The mean absolute error and the root mean square error of IKC are smaller than the other four algorithms, at the same dataset different sparsely, solving the data sparseness problem. IKC efficiency is also higher than the other four algorithms, on the same data set, solving the computationally intensive problems. So IKC improve the accuracy and timeliness won of recommender system.