The Spam Detection System Based On Mas Design And Realization

An Anti-Spam system called MASS(Multi-agent Anti-Spam System) has been introduced in this paper and MASS has been designed for EXCHANGE Emails Server and OUTLOOK in Windows environment. It could not only detect Spams that had ever been found before but automatically learn various Spams' characters which was labeled by users. Current Email software could supply simple filters ,but the filters need users to make complicated rules in several items. Recently, most of research about Anti-Spam have focused on Spam written in English and their design could not use in Spam written by Chinese because of the great difference between them. Popular Anti-Spam technology has been utilized in client and could not cooperate between Email Server and Client. Thus, MASS could get along with Spam written in English and Chinese and make Email Server and Client to cooperate harmoniously.The aim of MASS is to design and implement a system which makes Email Server and Client to cooperate harmoniously with distributed technology and AI, on the base of learning Anti-Spam technology and Emails essential concepts. During learning and practicing, many problems could raise and then be solved by us.At the beginning of the paper, concepts and developments about Anti-Spam has been introduced. Secondly, key technique and solution in system design and implement has been described in detail. Thirdly, this paper provides the architecture and implements methods in MASS. Finally, the problems ahead of Anti-Spam and some idea about them have been supposed.MASS is composed by four distributed agents: Server Agent, Client Agent, Control Agent, Subnet Agent and a central database. Server Agent would quickly , correctly detect Spams by the head of Emails and we use COM interfaces in EXCHANGE to implement it. Client Agent would quickly , correctly detect Spams by the text of Emails and this is made by using SVM( Support Vector Machine). Central database could store parameters usedby every user and personal rules so that administrator can query or set parameters in the feature. Control Agent could communicate with system administrator using graphical interface and it can provide results and parameter setting. Subnet Agent would detect Spam by subnet's characters when there are spams. Results of test imply MASS has satisfied correction.