Jiang et al. conducted a good study of SMS spam in a large cellular network in the US. They collected a large set of SMS spam through user-generated spam reports, which are more reliable and cleaner compared with other sources. After that, they tried to find SMS spam campaigns and anomalous behavior patterns through clustering based on URL and content similarity.

By comparing different campaigns, the authors obtained several interesting observations. For example, the top ten spam campaigns cover nearly half of the spam reports. They “are all related to cash advance[s], gift card[s], and free device[s] for testing.” And even the shortest campaign “lasts for more than one month, and half of the campaigns have a duration of approximately one year.” Finally, they proposed a simple two-step detection algorithm based on the correlation of spam numbers in terms of spatial and temporal features. The result shows that their approach has high accuracy and a very low false alarm rate.

This paper is very useful for spam detection researchers since it provides a first-hand observation of “the intentions and strategies of SMS spammers” and also shows the initial step in detecting SMS spam. We are in a mobile era now, and the number of mobile devices has reached more than three billion. Detecting SMS spam means an improvement in user experiences, and, more importantly, protection of users from information leaks, phishing attacks, and so on. This paper is a good starting point for large-scale SMS spam detection research.