Exploring Deceptive Domain Transfer Strategies: Mitigating the Differences among Deceptive Domains

Proceedings of Recent Advances in Natural Language Processing, pages 1076–1084 Varna, Sep 4–6, 2023 https://doi.org/10.26615/978-954-452-092-2_115 1076 Exploring Deceptive Domain Transfer Strategies: Mitigating the Differences among Deceptive Domains 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 Deceptive text poses a significant threat to users, resulting in widespread misinformation and disorder. While researchers have created numerous cutting-edge techniques for detecting deception in domain-specific settings, whether there is a generic deception pattern so that deception-related knowledge in one domain can be transferred to the other remains mostly unexplored. Moreover, the disparities in tex- tual expression across these many mediums pose an additional obstacle for generalization. To this end, we present a Multi-Task Learning (MTL) based deception generalization strategy to reduce the domain-specific noise and facil- itate a better understanding of deception via a generalized training. As deceptive domains, we use News (fake news), Tweets (rumors), and Reviews (fake reviews) and employ LSTM and BERT models to incorporate domain trans- fer techniques. Our proposed architecture for the combined approach of domain-independent and domain-specific training improves the de- ception detection performance by up to 5.28% in F1-score. 1 Introduction With the advent of the digital age came a deluge of textual content online, which also contains an enormous amount of deceptive text. The decep- tive text uses a variety of strategies to trick readers based on the information delivery medium. Fake news, for instance, spreads false information about a person or organization in order to harm their repu- tation, while fake reviews intentionally exaggerate the positive or negative aspects of a product or ser- vice in order to gain attention. Despite the technical differences in deception, all deceptive texts have the same objective of deceiving people; hence, a generic pattern of deception may exist (Shahriar et al., 2021). Identifying the underlying generic pat- tern of deception may unravel useful information about textual deception. Furthermore, such a sys- tem will enable a more effective detection approach through the intermingling of multiple deception do- mains. There has been a good number of work done to combat textual deception in domain-specific situa- tions, but a powerful detection system requires a lot of labeled data, which is dependent on things like trustworthy annotators, resources, time, and money. Consequently, the learning paradigm wherein mul- tiple domains may support each other can be a promising solution in deception detection. In this paper, we explore the feasibility of generalizing deception across domains such as News, Reviews, and Tweets, and we present a Multi-Task Learning (MTL) based deceptive domain transfer strategy to mitigate the domain differences and improve de- ception detection capacity beyond a standard single domain learning approach with LSTM and BERT models. Researchers dealt with deceptive domain adap- tation problems in cross-dataset learning settings, e.g., Opinion Spam on different entities (Li et al., 2014; S´anchez-Junquera et al., 2020), and Fake News on different topics (P´erez-Rosas et al., 2018). Although attempts have been made to …

Content truncated to 500 words