A Research On The Filter Algorithm Of Spam Information Based On One-class SVM

The research of monitoring and filtering of the files transportingthrough internet is getting hotter and hotter now. That information maycontain spam. The packets in network contain protocol parts. They can befragmented and encoded. Thus it is impossible for machine to recognizethe content of packets. And for content filtering, the traditional algorithmbased on string-matched is not able to meet the need of the huge increaseof information. Although SVM (Support vector machine) model can surelyimprove the efficiency of the classification, the problem that SVM’s toolarge dimension will affect the speed of examine still exists. It also causesa waste of storage space and compute ability.The paper will first analyze how to match the protocol automaticallyand then reassemble them to get the content information. It depends on theprotocol state machine. Based on that one algorithm is raised that by firstreducing the dimension by some specific algorithm before classification.The algorithm will speed up computation and save storage efficacy.The analysis result shows that after the improvement, a more accurateresult can be got. When even only selecting500feature values, theaccurate rate will be more than80%. And when selecting1000features,the accurate rate of DF algorithm will be more than90%.