Skip to content

Read-only version: For security reasons, the Research Information System (FIS) has been temporarily replaced by this read-only version. All public content and full texts remain available; login, editing, submission and statistics are not available. Questions: fis@uni-bamberg.de

Home › Publications › Hierarchical Adversarial Correction to Mitigate Identity Term Bias in Toxicity Detection

Publication

Hierarchical Adversarial Correction to Mitigate Identity Term Bias in Toxicity Detection

Schäfer, Johannes; Heid, Ulrich; Klinger, Roman (2025): Hierarchical Adversarial Correction to Mitigate Identity Term Bias in Toxicity Detection, in: Bamberg: Otto-Friedrich-Universität, S. 35–51.

Details

Faculty/Chair
Faculty/Professorship
Author
Schäfer, Johannes
Heid, Ulrich
Publisher Information
Otto-Friedrich-Universität : Bamberg
Year of publication
2025
Source/Other editions
Orphee De Clercq, Valentin Barriere, Jeremy Barnes, u. a. (Hrsg.), Proceedings of the 14th Workshop on Computational Approaches to Subjectivity, Sentiment, & Social Media Analysis, Bangkok, Thailand: Association for Computational Linguistics, 2024, S. 35–51
Year of first publication
2024
Language
English
Licence
URN
Abstract

Corpora that are the fundament for toxicity detection contain such expressions typically directed against a target individual or group, e.g., people of a specific gender or ethnicity. Prior work has shown that the target identity mention can constitute a confounding variable. As an example, a model might learn that Christians are always mentioned in the context of hate speech. This misguided focus can lead to a limited generalization to newly emerging targets that are not found in the training data. In this paper, we hypothesize and subsequently show that this issue can be mitigated by considering targets on different levels of specificity. We distinguish levels of (1) the existence of a target, (2) a class (e.g., that the target is a religious group), or (3) a specific target group (e.g., Christians or Muslims). We define a target label hierarchy based on these three levels and then exploit this hierarchy in an adversarial correction for the lowest level (i.e. (3)) while maintaining some basic target features. This approach does not lower the toxicity detection performance but increases the generalization to targets not being available at training time.

Keywords
Hierarchical Adversarial Correction
Peer Reviewed
yes
International Distribution
yes
Open Access Journal
yes
Type
Conferenceobject
URI
Activation date
2025-11-10

Permalink: https://fis.uni-bamberg.de/handle/uniba/111086 · Data as JSON · Cite: BibTeX · RIS

Data as of: 2026-11-16

Restricted read-only access: login, editing and submission are temporarily disabled.