Research On E-commerce Recommendation Algorithm Based On Singular Value Decomposition And K-means Clustering

With the massive coverage of the Internet and the rapid development of e-commerce, the expanding network of information and network resources, takes the user to an era of information overload.With the sustained growth of the mass of information, network users can not quickly find the part their real needs..How to active positioning for the user from the vast amounts of resources and push the content that may be of interest to is the main task of the current recommendation system. The biggest advantage of the recommendation system that can quickly push resource users are really interested in, alleviating the pressure of information retrieva.At present, in numerous recommendation technology, collaborative filtering recommendation is the most successful and most widely application recommendation technology.Research in the field of collaborative filtering, the model problems mainly related mostly is the data sparseness and scalability, as well as the accuracy of the data recommended problem, and k-means clustering algorithm is an unsupervised learning algorithm has good classification performance, and a wide range of applications, concise algorithm. However, with the data matrix sparsity, the clustering algorithm can effectively be allocated in accordance with similar interests hobby users to the same cluster; after clustering, according to neighbors users predict evaluation target for the goods. But the existence of the defect is due to sparse data, is the target audience at the edge of clustering, the lower the accuracy of the target user’s recommendation, in view of this, this paper uses the singular value decomposition technique to resolve the sparsity of data caused by the clustering of data objects adverse effects; then the randomness of the initial cluster centers for the improved algorithm proposed fusion of collaborative filtering algorithm and its simulation experiments show that the proposed algorithm has good performance recommendation.The basic idea of the singular value decomposition based recommendation algorithm:First to collaborative filtering algorithm as the main basis for the recommendation accuracy appear sparse data problems caused by low, using the singular value decomposition of the scoring matrix dimension reduction, and combined with the gradient descent method for users to update and project characteristics, effectively avoiding the overfitting phenomenon, while overcoming a zero score for the user appears recommend not fine when calculating the similarity problem. MovieLence datasets by testing and recommendation algorithm with traditional recommendation when comparing the results show that the algorithm had a better accuracy in the recommendation.An improved initial centroid of the basic idea of k-means clustering algorithm are: the randomness of the traditional k-means clustering algorithm k initial cluster centers choice, resulting in volatility clustering results, we propose a be able to generate more stable improved algorithm initial cluster centers; at the same time, with the equalization function effectively find the best k value. Through simulation experiments UCI datasets, and with the traditional k-means clustering algorithm are compared, the results show that the improved algorithm has better performance.The basic idea of model-based recommendation algorithm KS VD is:appears when introducing improved k-means clustering algorithm to classify data objects to handle high-dimensional sparse data on the negative impact caused by the algorithm, using singular value decomposition technique, on the other space messaging re-expression data. By MovieLence dataset function simulation and comparison with other recommendation algorithm, the results show that the improved algorithm has better search capability.