TurQUaz at CheckThat! 2024: Creating Adversarial Examples using Genetic Algorithm

Abstract

As we increasingly integrate artificial intelligence into our daily tasks, it is crucial to ensure that these systems are reliable and robust against adversarial attacks. In this paper, we present our participation in Task 6 of CLEF CheckThat! 2024 lab. In our work, we explore several methods, which can be grouped into two categories. The first group focuses on using a genetic algorithm to detect words and changing them via several methods such as adding/deleting words and using homoglyphs. In the second group of methods, we use large language models to generate adversarial attacks. Based on our comprehensive experiments, we pick the genetic algorithm-based model which utilizes a combination of splitting words and homoglyphs as a text manipulation method, as our primary model. We are ranked third based on both BODEGA metric and manual evaluation. © 2024 Copyright for this paper by its authors.