Style Transfer as Bias Mitigation: Diffusion Models for Synthetic Mental Health Text for Arabic

2026-01-20

AI Summary

This work addresses the challenges of gender imbalance and data scarcity in Arabic mental health texts, where existing large language model–based synthesis methods often exacerbate biases and lack diversity. The study introduces, for the first time, a diffusion-based approach to text style transfer in this low-resource domain, proposing a framework that operates without pretraining to transform male-authored texts into female-styled ones using the CARMA corpus. By constructing five subsets that capture distinct linguistic manifestations of gender expression in Arabic, the model effectively disentangles semantic content from surface stylistic features. This enables the generation of high-entropy, unbiased female-style texts that preserve strong semantic fidelity. Experimental results demonstrate that the generated outputs exhibit both linguistic plausibility and pronounced gender-specific stylistic differences, offering a novel pathway to mitigate data bias in sensitive domains.

Technology Category

• Natural Language Processing: Sentiment Analysis, Stylistic Analysis, and Argument Mining
• Machine Learning: Large Multimodal Models (LMMs)
• Computer Vision: Diffusion Models for Vision

Application Category

• Web Mining and Content Analysis: Large pretrained models with web data
• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Social Networks and Social Media: Generative AI / large language models and their impact on social systems

Abstract

Synthetic data offers a promising solution for mitigating data scarcity and demographic bias in mental health analysis, yet existing approaches largely rely on pretrained large language models (LLMs), which may suffer from limited output diversity and propagate biases inherited from their training data. In this work, we propose a pretraining-free diffusion-based approach for synthetic text generation that frames bias mitigation as a style transfer problem. Using the CARMA Arabic mental health corpus, which exhibits a substantial gender imbalance, we focus on male-to-female style transfer to augment underrepresented female-authored content. We construct five datasets capturing varying linguistic and semantic aspects of gender expression in Arabic and train separate diffusion models for each setting. Quantitative evaluations demonstrate consistently high semantic fidelity between source and generated text, alongside meaningful surface-level stylistic divergence, while qualitative analysis confirms linguistically plausible gender transformations. Our results show that diffusion-based style transfer can generate high-entropy, semantically faithful synthetic data without reliance on pretrained LLMs, providing an effective and flexible framework for mitigating gender bias in sensitive, low-resource mental health domains.

Problem

• bias mitigation
• gender imbalance
• synthetic data
• mental health
• Arabic

Innovation

• diffusion models
• style transfer
• bias mitigation
• synthetic text generation
• Arabic mental health