A Domain-Independent Holistic Approach to Deception Detection

Proceedings of Recent Advances in Natural Language Processing, pages 1308–1317 Sep 1–3, 2021. https://doi.org/10.26615/978-954-452-072-4_147 1308 A Domain-Independent Holistic Approach to Deception Detection Sadat Shahriar University of Houston, Texas, USA sshahriar@uh.edu Arjun Mukherjee University of Houston, Texas, USA arjun@cs.uh.edu Omprakash Gnawali University of Houston, Texas, USA gnawali@cs.uh.edu Abstract The deception in the text can be of differ- ent forms in different domains, including fake news, rumor tweets, and spam emails. Ir- respective of the domain, the main intent of the deceptive text is to deceit the reader. Al- though domain-specific deception detection exists, domain-independent deception detec- tion can provide a holistic picture, which can be crucial to understand how deception oc- curs in the text. In this paper, we detect de- ception in a domain-independent setting using deep learning architectures. Our method out- performs the State-of-the-Art (SOTA) perfor- mance of most benchmark datasets with an overall accuracy of 93.42% and F1-Score of 93.22%. The domain-independent training al- lows us to capture subtler nuances of decep- tive writing style. Furthermore, we analyze how much in-domain data may be helpful to accurately detect deception, especially for the cases where data may not be readily available to train. Our results and analysis indicate that there may be a universal pattern of deception lying in-between the text independent of the domain, which can create a novel area of re- search and open up new avenues in the field of deception detection. 1 Introduction In the current era of the flood of information, de- ception has become an undeniable event, causing a financial or political catastrophe and even the loss of human lives. Often, we do not have the neces- sary resources to validate a tweet or a catchy social media forwarded news link. Our idea is to capture the writing style hidden “between-the-lines”, in- tended to deceive the reader. The deception can be in the form of any textual stream and on any topic. So, our objective is to find a holistic model that can leverage thousands of textual resources and find a learning architecture to decode the deception. Adopting the definition of Deception from Bur- goon and Buller, we define Deceptive Text as any textual content that aims to misconstrue an affair in a deliberate way causing the reader at a disad- vantage either directly or indirectly (Burgoon and Buller, 1994). Deceptive text can be of various forms. For example, in the news and public me- dia domain, the deceptive text is known as Fake News. In social media, a deceptive text can hap- pen in the form of a rumor. Spam or a phishing email is treated as deceptive content in the per- sonal mail or messaging domain. Each domain’s deceptive text has distinct ways to deceit the reader. While fake news can spread falsified propaganda, spam or phishing email can be used for merely monetary gain. Therefore, the ways of formulat- ing a deceptive text can have significant variations. Notwithstanding, all deceptive texts share a com- mon goal of tricking the reader …

Content truncated to 500 words