Identity Deception Detection

Proceedings of the The 8th International Joint Conference on Natural Language Processing, pages 885–894, Taipei, Taiwan, November 27 – December 1, 2017 c⃝2017 AFNLP Identity Deception Detection Ver´onica P´erez-Rosas1, Quincy Davenport1, Anna Mengdan Dai1, Mohamed Aboulenien2 and Rada Mihalcea1 1Computer Science and Electrical Engineering. University of Michigan 2Computer and Information Science. University of Michigan, Dearborn {vrncapr,quincyd,annadai,zmohamed,mihalcea}@umich.edu Abstract This paper addresses the task of detecting identity deception in language. Using a novel identity deception dataset, consist- ing of real and portrayed identities from 600 individuals, we show that we can build accurate identity detectors targeting both age and gender, with accuracies of up to 88%. We also perform an analysis of the linguistic patterns used in identity decep- tion, which lead to interesting insights into identity portrayers. 1 Introduction With the ever growing usage of social media and other online interactions, cyber-crimes such as identity thief, fraud, and sexual predation have be- come increasingly common. The availability of a wide variety of social platforms and apps fur- ther facilitates these kind of crimes, which are often characterized by the ease of deception and concealment of one’s real identity. Moreover, the increased sense of security due to the spatial and temporal “distanciation” involved with online communications leads to a growing number of oc- currences of identity deception. Identity deception is defined as “pretending to be someone you are not” (Rowe, 2009). The inten- tions of identity deceivers differ between achiev- ing monetary benefits, committing fraud, sexual predating, appearing more attractive in online dat- ing, and so on. The risk is further increased with the massive number of children and teens using social media as well as the elderly who get in- volved in online interactions that they assume to be trustworthy by default due to their lack of ex- perience. Multiple stories of teenagers who be- came victims of these activities, as well as elderly who lost hundreds of thousands of dollars are now often encountered in the news. For instance, a re- cent study reported that 92% of teens go online on daily basis including 24% who are online “almost constantly”. 1 The study additionally reported that females tend to use social media more than males. Hence, due to the financial, social, physical, and psychological damages associated with cyber- crimes, there is a growing need to learn more about the patterns and behaviors of identity deceivers, in order to develop computational approaches that can aid in preventing such crimes. In this work, we seek to identify linguistic dif- ferences in individuals’ self-presentation when en- gaging in identity deception behaviors. We target the two most common behaviors related with on- line identity deception i.e., individuals portraying themselves as either younger and older, and indi- viduals lying about their gender –also known as gender switching (Macwan and inz. Grzegorz Fil- cek, 2017; Herring and Martinson, 2004) . We perform a set of experiments to explore three main research questions. First, given a de- ceptive corpus consisting of written samples of gender switching, can …

Content truncated to 500 words