English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

The advent of social media has given rise to numerous ethical challenges,with hate speech among the most significant concerns. Researchers areattempting to tackle this problem by leveraging hate-speech detection andemploying language models to automatically moderate content and promote civildiscourse. Unfortunately, recent studies have revealed that hate-speechdetection systems can be misled by adversarial attacks, raising concerns abouttheir resilience. While previous research has separately addressed therobustness of these models under adversarial attacks and theirinterpretability, there has been no comprehensive study exploring theirintersection. The novelty of our work lies in combining these two criticalaspects, leveraging interpretability to identify potential vulnerabilities andenabling the design of targeted adversarial attacks. We present a comprehensiveand comparative analysis of adversarial robustness exhibited by varioushate-speech detection models. Our study evaluates the resilience of thesemodels against adversarial attacks using explainability techniques. To gaininsights into the models' decision-making processes, we employ the LocalInterpretable Model-agnostic Explanations (LIME) framework. Based on theexplainability results obtained by LIME, we devise and execute targeted attackson the text by leveraging the TextAttack tool. Our findings enhance theunderstanding of the vulnerabilities and strengths exhibited bystate-of-the-art hate-speech detection models. This work underscores theimportance of incorporating explainability in the development and evaluation ofsuch models to enhance their resilience against adversarial attacks.Ultimately, this work paves the way for creating more robust and reliablehate-speech detection systems, fostering safer online environments andpromoting ethical discourse on social media platforms.

Exploiting Explainability to Design Adversarial Attacks and Evaluate Attack Resilience in Hate-Speech Detection Models