Multi-standard Reputation-model-based Detection System Of Spam Call

With the development of communication technology, mobile phone users areincreasing rapidly. Although people enjoy the convenience of this technology, cell phonegradually become the target of spam calls because of its personal and privacy feature.According to survey, more than90%people are exposed to the harassment of spam calls,the surfeit of spam calls are annoying people’s daily life while wasting much bandwidthresource. On the condition that we lack enough bylaws to impose sanctions to thosespam-call-senders, it’s a better choice to use technological methods to weaken those spamcalls.There are many methods have been proposed to detect spam-calls, such asblack-and-white list, Turing test, prepaid fees, these methods have already shown someeffects. But more and more spam-calls are human issue, they can bypass the machinedetection; they are too cunning to detect. So people begin to focus on the reputationsystem. The research on reputation system have already achieved good results, but thereare still some issues haven’t been taken into account:1) How to deal with multi-standardcriteria?2) How to accelerate learning speed?3) How to make filter model more flexiblewhen users’ behavior is changing?To solve those problems above, this paper proposes a multi-standard reputation modelbased on group learning solution. The user to be classified according to their disposition totrust solved the multi-model evaluation criteria problem; group learning mechanism hasbeen established after dividing users into different small groups, so the model processingspeed will not slow down because of the large amounts of data; besides, with the streamdata processing methods and timing of the sliding window model update mechanismensures that the model is able to change according to the user’s changes to ensure theaccuracy of the model.Simulation results show that the introduction of dividing trust tendency makes thejudgment result of the model, compared to ordinary reputation models, have betterdetection effect; with the introduction of group learning mechanism, learning rateimproved significantly even if each node in the condition of low transaction rates. Themodel meets the design requirements very well.