Misinformation Detection using Large Language Models with Explainability

2025-10-21

AI Summary

To address the erosion of trust and impaired decision-making caused by misinformation propagation on online platforms, this paper proposes an interpretable and lightweight misinformation detection framework. Methodologically: (1) it adopts DistilBERT as the backbone, enhanced by hierarchical unfreezing and layer-wise learning rate decay to improve fine-tuning efficiency; (2) it introduces a dual-level interpretability mechanism integrating LIME (token-level local explanations) and SHAP (model-level global attribution), balancing transparency and reliability. Evaluated on the COVID Fake News and FakeNewsNet GossipCop datasets, the framework achieves accuracy comparable to RoBERTa while accelerating inference by 40% and significantly reducing computational overhead. Its primary contribution lies in the first synergistic design of a lightweight pre-trained model with dual-granularity interpretability techniques—unifying high accuracy, computational efficiency, and strong interpretability—making it suitable for scalable, trustworthy content moderation in large-scale social media and news platforms.

Technology Category

• Natural Language Processing: Fact-Checking / Misinformation Detection (NLP Focus)
• Application Domains: Misinformation & Fake News
• Machine Learning: Calibration & Uncertainty Quantification

Application Category

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

Abstract

The rapid spread of misinformation on online platforms undermines trust among individuals and hinders informed decision making. This paper shows an explainable and computationally efficient pipeline to detect misinformation using transformer-based pretrained language models (PLMs). We optimize both RoBERTa and DistilBERT using a two-step strategy: first, we freeze the backbone and train only the classification head; then, we progressively unfreeze the backbone layers while applying layer-wise learning rate decay. On two real-world benchmark datasets, COVID Fake News and FakeNewsNet GossipCop, we test the proposed approach with a unified protocol of preprocessing and stratified splits. To ensure transparency, we integrate the Local Interpretable Model-Agnostic Explanations (LIME) at the token level to present token-level rationales and SHapley Additive exPlanations (SHAP) at the global feature attribution level. It demonstrates that DistilBERT achieves accuracy comparable to RoBERTa while requiring significantly less computational resources. This work makes two key contributions: (1) it quantitatively shows that a lightweight PLM can maintain task performance while substantially reducing computational cost, and (2) it presents an explainable pipeline that retrieves faithful local and global justifications without compromising performance. The results suggest that PLMs combined with principled fine-tuning and interpretability can be an effective framework for scalable, trustworthy misinformation detection.

Problem

• Detecting misinformation using explainable transformer-based language models
• Optimizing computational efficiency while maintaining detection accuracy
• Providing transparent local and global explanations for model predictions

Innovation

• Two-step fine-tuning with layer-wise learning rate decay
• Integrating LIME and SHAP for multi-level explainability
• Lightweight DistilBERT achieves comparable accuracy to RoBERTa