Description

Short Message Service (SMS) text messages are indispensable in our lives today, but along with the convenience of using SMS messages in our daily lives, we also face a serious problem caused by SMS spamming. In this article, we explore a service-side solution to use graph data mining in finding out spammers from non-spammers. An important issue is whether we can detect spammers without checking into the contents of the messages; these are the content-less features. Using a real-world data set from a large telecommunications operator in China, we examine the effectiveness of various content-less features that range from network and to time-oriented categories. We find that some intuitively appealing features are in fact not very effective, whereas a combination of temporal and network features can be very useful in training high-performance classifiers for spammer detection