Voice Spam Call Behavior Analyses Based On Data Stream

In recent years, with the development of information technology on global society, thereal-time communication has become an important part of people’s daily lives.With thecombination of PSTN, mobile network and Internet, voice applications have been widelysupplied. However, because of the impact of spam voice, the development of voice servicehas been challenged. Except for the security of users’ communication, spam voice wastedresources and bandwidth of communication, even affect social stability in worst cases.When normal users communicate normally, spam callers should be limited at the sametime.According to spam calls, there are many methods have been proposed. The call modelanalysis has been identified effective without affecting users’ communication. Existingcall models can not dynamically reflect the characteristics of the user. It is easy for spamvoice to changing behavior and avoiding detecting system. While spammers change callbehavior, detecting system can not adaptive response to the modification and it need to beimproved.With the combination of the characteristics of users’ call behavior analysis, the spamvoice filtering mechanism is designed based on data stream clustering. Call records havethree parameters: callers, call time and duration. After introduce users’ daily life habits,proposed call behavior model is composed of call interaction, call frequency withdistribution and call duration with distribution. It is difficult for spammers to disguise thecall behavior. Data similarity funcation is classified to distance and similarity coefficientaccording to the call data characteristics. Streaming data algorithms can extracte datafeature and clusters users. Dynamically output of the stream data algorithm can adapt usercalling behavior changing. Finally, the nearest neighbor classification is used to detectwhether the user is a spammer.From the simulation analysis, combinition of variety of characteristics obviouslyshows higer detectiong result. Data similarity funcation improved the detection efficiencysignificantly. Stream data algorithm has lower processing time without lower detectionperformance. While user calling behavior changing, stream data algorithm analysis callbehavior dynamically and work well in detecting system.