Face the Facts! Evaluating RAG-based Fact-checking Pipelines in Realistic Settings

2024-12-19

arXiv.org

AI Summary

This study systematically evaluates retrieval-augmented generation (RAG) for automated fact-checking in realistic settings, focusing on truth adjudication of stylistically complex claims and generation of concise verdicts, while assessing robustness across heterogeneous yet trustworthy knowledge bases. Method: We propose the first end-to-end framework integrating LLM-driven retrieval, multi-scale language models (zero-/few-shot and fine-tuned), and hybrid human-automated evaluation. Benchmarking is conducted across four dimensions: faithfulness, context adherence, sentiment alignment, and informativeness. Contribution/Results: Key findings reveal that LLM-based retrievers substantially outperform traditional methods but remain sensitive to knowledge heterogeneity; larger LMs improve adjudication faithfulness, whereas smaller LMs excel in context adherence; zero-/few-shot generators yield more informative verdicts, while fine-tuned models better capture sentiment nuance. The results expose complementary trade-offs between model scale and specialized capabilities, offering principled guidance for RAG deployment in fact-checking.

Technology Category

• Natural Language Processing: Fact-Checking / Misinformation Detection (NLP Focus)
• Machine Learning: Large Multimodal Models (LMMs)
• Data Mining & Knowledge Management: Conversational Systems for Recommendation & Retrieval

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Natural Language Processing and Generation systems have recently shown the potential to complement and streamline the costly and time-consuming job of professional fact-checkers. In this work, we lift several constraints of current state-of-the-art pipelines for automated fact-checking based on the Retrieval-Augmented Generation (RAG) paradigm. Our goal is to benchmark, under more realistic scenarios, RAG-based methods for the generation of verdicts - i.e., short texts discussing the veracity of a claim - evaluating them on stylistically complex claims and heterogeneous, yet reliable, knowledge bases. Our findings show a complex landscape, where, for example, LLM-based retrievers outperform other retrieval techniques, though they still struggle with heterogeneous knowledge bases; larger models excel in verdict faithfulness, while smaller models provide better context adherence, with human evaluations favouring zero-shot and one-shot approaches for informativeness, and fine-tuned models for emotional alignment.

Problem

• Evaluating RAG-based fact-checking pipelines in realistic settings
• Benchmarking verdict generation on complex claims and heterogeneous knowledge
• Assessing performance trade-offs between different model configurations

Innovation

• RAG-based automated fact-checking pipeline
• Evaluating LLM retrievers on complex claims
• Benchmarking verdict generation with heterogeneous knowledge bases