Improved SVM Method Applied to the Online User Behavior Analysis

摘要

Online user behavior analysis has gained extensive attention in recent years. In this paper, to obtain the real users' online behaviors based on a DNS-level tracing approach, a new improved SVM (support vector machine) method for analyzing the users' online behaviors is put forth, which enables to get insightful views at a large scale. As the increase of the amount of data, improving the convergence speed of SVM is highly desired. The computational efficiency of the proposed SVM of this work is greatly improved by rewriting KKT conditions for the Sequential Minimal Optimization (SMO) algorithm. The improved SVM possesses a great capability of clustering the users' data and revealing the users' behaviors accurately from various aspects. The effectiveness of the improved SVM method is validated and demonstrated via analyzing a set of data of users' online behaviors.