TextGuard: Identifying and Neutralizing Adversarial Threats in Textual Data

Albtsoh, Luay
Omar, Marwan

International Journal of Informatics, Information System and Computer Engineering (INJIISCOM)
Vol. 6 No. 2 (2025): INJIISCOM: VOLUME 6, ISSUE 2, DECEMBER 2025 (ONLINE FIRST)

Adversarial attacks inside the text domain pose a serious risk to the integrity of Natural Language Processing (NLP) systems. In this study, we propose "Text-Guard," a unique approach to detect hostile instances in natural language processing, based on the Local Outlier Factor (LOF) algorithm. This paper compares TextGuard's performance against that of more traditional NLP classifiers such as LSTM, CNN, and transformer-based models, while also experimentally verifying its effectiveness on a variety of real-world datasets. TextGuard significantly surpasses earlier state-of-the-art methods like DISP and FGWS, with F1 recognition accuracy scores as high as 94.8%. This sets a new benchmark in the field as the first use of the LOF technique for adversarial example identification in the text domain