HEARTS: A Holistic Framework for Explainable, Sustainable and Robust Text Stereotype Detection

TL;DR

HEARTS is a comprehensive framework that improves stereotype detection in text by combining high performance, explainability, and sustainability, using a new diverse dataset and interpretability techniques.

Contribution

The paper introduces HEARTS, a novel holistic framework for stereotype detection that integrates a new dataset, model performance, explainability, and environmental sustainability.

Findings

Fine-tuned BERT models outperform individual components.

ALBERT-V2 with SHAP provides human-aligned explanations.

HEARTS reduces carbon footprint while maintaining accuracy.

Abstract

Stereotypes are generalised assumptions about societal groups, and even state-of-the-art LLMs using in-context learning struggle to identify them accurately. Due to the subjective nature of stereotypes, where what constitutes a stereotype can vary widely depending on cultural, social, and individual perspectives, robust explainability is crucial. Explainable models ensure that these nuanced judgments can be understood and validated by human users, promoting trust and accountability. We address these challenges by introducing HEARTS (Holistic Framework for Explainable, Sustainable, and Robust Text Stereotype Detection), a framework that enhances model performance, minimises carbon footprint, and provides transparent, interpretable explanations. We establish the Expanded Multi-Grain Stereotype Dataset (EMGSD), comprising 57,201 labelled texts across six groups, including under-represented demographics like LGBTQ+ and regional stereotypes. Ablation studies confirm that BERT models fine-tuned on EMGSD outperform those trained on individual components. We then analyse a fine-tuned, carbon-efficient ALBERT-V2 model using SHAP to generate token-level importance values, ensuring alignment with human understanding, and calculate explainability confidence scores by comparing SHAP and LIME outputs...