Explaining Veracity Predictions with Evidence Summarization: A Multi-Task Model Approach

TL;DR

This work addresses automated misinformation detection by jointly learning veracity prediction and evidence-based explanations. It introduces a multi-task framework that uses a shared T5 encoder and a dedicated summarization decoder, with veracity classified by side heads, trained under a combined loss $Loss = w_s*Loss_{summ} + w_c*Loss_{cl}$ and with static or uncertainty-based weighting. Evaluations on PUBHEALTH, FEVER, and e-FEVER show that multi-task training can improve summarization quality or veracity accuracy depending on the base model (T5 vs Flan-T5), with evidence retrieval and data-imbalance factors influencing results. The approach advances interpretable fact-checking by integrating explanation generation directly into the model’s learning objective, potentially aiding users and downstream systems in understanding and trusting veracity judgments. Future work includes user studies on explanation coherence and exploring parameter-efficient fine-tuning techniques like LoRA for broader applicability.

Abstract

The rapid dissemination of misinformation through social media increased the importance of automated fact-checking. Furthermore, studies on what deep neural models pay attention to when making predictions have increased in recent years. While significant progress has been made in this field, it has not yet reached a level of reasoning comparable to human reasoning. To address these gaps, we propose a multi-task explainable neural model for misinformation detection. Specifically, this work formulates an explanation generation process of the model's veracity prediction as a text summarization problem. Additionally, the performance of the proposed model is discussed on publicly available datasets and the findings are evaluated with related studies.

Figures (2)

• Figure 1: The multi-task model architecture
• Figure 2: A sample claim from PUBHEALTH kotonya2020explainable with our model's outputs