Anti-spit Technology Research Based On Support Vector Machine

With the development of the next-generation network technology, the gradual integration of the Internet and the traditional telecommunications network has become the inevitable trend of network development, and the VoIP technology caters to this trend. As an important VoIP control signaling, SIP(Session Initiation Protocol) has been identified as the core protocol of the third generation mobile communication network by 3GPP(3rd Generation Partnership Project).But the security problem especial the SPIT(Spam over Internet Telephony) over the VoIP network has become an important issue with the development and extensive application of VoIP. In 2008, IETF(The Internet Engineering Task Force) posed RFC 5039 which deeply analyzes the problem of spam in SIP.This dissertation researches SIP and Anti-SPIT technology and deeply analyzes the feature of spam call behavior. In this dissertation, Support Vector Machine and the application of Support Vector Machine in Anti-SPIT technology are researched. The main work of this dissertation as follows:1) Preprocessing Algorithm ResearchAttribute number and sample size of training sample set are important factors which affect the training speed. This dissertation researches attributes reduction algorithm based on discernibility matrix and finds that the Support Vector Machine with a combination of attributes reduction algorithm can reduce the training time. K-NN (K-Nearest Neighbor) algorithm and the application of the K-NN in data preprocessing are also researched in this dissertation.2) Training Algorithm ResearchThis dissertation researches the training algorithm of Support Vector Machine and deeply analyzes the factors such as step extrapolation parameter, cache and parallel computing which affect the performance of SMO algorithm. Mention to the sample which violates the KKT condition updated only once, an improved SMO algorithm is proposed, which choose another sample with the same sample volates the KKT condition for the second update in the case of the initial update failed. The improved algorithm has a higher training precision but a longer training time.3) Research on the application of Support Vector Machine in Anti-SPIT technologyThis dissertation researches SIP and Anti-SPIT technology and deeply analyzes the feature of spam call behavior. An application layer SIP security gateway is designed and the implementation of naomichi filter module and Turing Test module are described in this dissertation. In the research on the usage of Support Vector Machine, a data record format used to store the user call information is designed. Through the record format, awk can be effective used to extract features of users call behavior. Through the analysis of updated samples set, a data training strategie using KKT conditions to select new samples is proposed.