Improving Hate Speech Classification with Cross-Taxonomy Dataset Integration

2025-03-07

AI Summary

Hate speech detection suffers from poor generalizability due to heterogeneous definitions and dataset annotations across domains. Method: This paper introduces the first universal cross-taxonomy framework for semantic alignment and joint modeling of multi-source annotation standards (e.g., HateSpeech18, FDCL19). We propose a BERT-based multitask learning architecture integrating terminology mapping, label-space alignment, and adversarial domain adaptation, enabling a single model to jointly classify hate speech subtypes under legal, platform-specific, and research-oriented definitions. Contribution/Results: Our approach achieves a 6.2-point F1-score improvement on independent test sets and an average 11.4% gain in cross-definition transfer accuracy. It significantly reduces reliance on multiple specialized models, offering a scalable, interpretable, and unified modeling paradigm for hate speech identification.

Technology Category

• Machine Learning: Transfer, Domain Adaptation, Multi-Task Learning
• Natural Language Processing: Machine Translation, Multilinguality, Cross-Lingual NLP
• Application Domains: Humanities & Computational Social Science

Application Category

• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Web Mining and Content Analysis: Mining multimedia, multimodal, multilingual, cross-lingual Web data
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Algorithmic hate speech detection faces significant challenges due to the diverse definitions and datasets used in research and practice. Social media platforms, legal frameworks, and institutions each apply distinct yet overlapping definitions, complicating classification efforts. This study addresses these challenges by demonstrating that existing datasets and taxonomies can be integrated into a unified model, enhancing prediction performance and reducing reliance on multiple specialized classifiers. The work introduces a universal taxonomy and a hate speech classifier capable of detecting a wide range of definitions within a single framework. Our approach is validated by combining two widely used but differently annotated datasets, showing improved classification performance on an independent test set. This work highlights the potential of dataset and taxonomy integration in advancing hate speech detection, increasing efficiency, and ensuring broader applicability across contexts.

Problem

• Integrates diverse hate speech datasets into a unified model.
• Proposes a universal taxonomy for broader hate speech detection.
• Enhances classification performance across different contexts.

Innovation

• Integrated cross-taxonomy datasets for unified model
• Developed universal taxonomy for hate speech classification
• Enhanced classifier performance with dataset integration