From Fragments to Facts: A Curriculum-Driven DPO Approach for Generating Hindi News Veracity Explanations

2025-07-07

AI Summary

Addressing the challenge of generating faithful, interpretable explanations for news veracity in low-resource languages—particularly Hindi—where automated tools remain scarce, this paper proposes a novel framework integrating Direct Preference Optimization (DPO) with curriculum learning. Methodologically, it introduces two orthogonal alignment metrics—“Actuality” (ensuring factual consistency) and “Finesse” (capturing explanatory nuance)—into the DPO loss function. The approach leverages multilingual foundation models (e.g., Mistral, Llama, Gemma) and sequence-to-sequence models (e.g., mBART, mT5), fine-tuned on Hindi misinformation data. Empirical results demonstrate substantial improvements in explanation coherence, contextual relevance, and credibility over strong baselines. Quantitatively, the framework achieves significant gains in explanation accuracy and human evaluation scores. This work establishes a scalable, high-fidelity paradigm for automated, linguistically grounded explanation generation in low-resource settings, advancing misinformation detection for under-resourced languages.

Technology Category

• Natural Language Processing: Fact-Checking / Misinformation Detection (NLP Focus)
• Humans and AI: Explainable AI (XAI) for Human Understanding
• Application Domains: Misinformation & Fake News

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Web Mining and Content Analysis: Large pretrained models with web data
• Responsible Web: Mitigating misinformation and disinformation

Abstract

In an era of rampant misinformation, generating reliable news explanations is vital, especially for under-represented languages like Hindi. Lacking robust automated tools, Hindi faces challenges in scaling misinformation detection. To bridge this gap, we propose a novel framework integrating Direct Preference Optimization (DPO) with curriculum learning to align machine-generated explanations with human reasoning. Fact-checked explanations from credible sources serve as preferred responses, while LLM outputs highlight system limitations and serve as non-preferred responses. To refine task-specific alignment, we introduce two key parameters -- Actuality and Finesse -- into the DPO loss function, enhancing explanation quality and consistency. Experiments with LLMs (Mistral, Llama, Gemma) and PLMs (mBART, mT5) confirm the framework's effectiveness in generating coherent, contextually relevant explanations. This scalable approach combats misinformation and extends automated explanation generation to low-resource languages.

Problem

• Generating reliable Hindi news explanations to combat misinformation
• Scaling misinformation detection for under-represented languages like Hindi
• Aligning machine-generated explanations with human reasoning using DPO

Innovation

• Curriculum-driven DPO for Hindi news veracity
• Enhanced DPO loss with Actuality and Finesse
• Scalable framework for low-resource language explanations