An analysis of scam baiting calls: Identifying and extracting scam stages and scripts

Phone scams remain a difficult problem to tackle due to the combination of protocol limitations, legal enforcement challenges and advances in technology enabling attackers to hide their identities and reduce costs. Scammers use social engineering techniques to manipulate victims into revealing their personal details, purchasing online vouchers or transferring funds, causing significant financial losses. This paper aims to establish a methodology with which to semi-automatically analyze scam calls and infer information about scammers, their scams and their strategies at scale. Obtaining data for the study of scam calls is challenging, as true scam victims do not in general record their conversations. Instead, we draw from the community of ``scam baiters'' on YouTube: individuals who interact knowingly with phone scammers and publicly publish their conversations. These can not be considered as true scam calls, however they do provide a valuable opportunity to study scammer scripts and techniques, as the scammers are unaware that they are not speaking to a true scam victim for the bulk of the call. We applied topic and time series modeling alongside emotion recognition to scammer utterances and found clear evidence of scripted scam progressions that matched our expectations from close reading. We identified social engineering techniques associated with identified script stages including the apparent use of emotion as a social engineering tool. Our analyses provide new insights into strategies used by scammers and presents an effective methodology to infer such at scale. This work serves as a first step in building a better understanding of phone scam techniques, forming the ground work for more effective detection and prevention mechanisms that draw on a deeper understanding of the phone scam phenomenon.