A Comparative Analysis of Faithfulness Metrics and Humans in Citation Evaluation

TL;DR

This paper evaluates the effectiveness of faithfulness metrics in distinguishing different levels of citation support in LLM-generated content, revealing current metrics' limitations and guiding future improvements.

Contribution

It introduces a comprehensive framework for assessing faithfulness metrics in fine-grained citation support scenarios, highlighting their inconsistent performance across support levels.

Findings

No single metric outperforms others across all evaluations.

Current metrics struggle to differentiate partial support from full or no support.

The evaluation framework provides a more nuanced understanding of metric effectiveness.

Abstract

Large language models (LLMs) often generate content with unsupported or unverifiable content, known as "hallucinations." To address this, retrieval-augmented LLMs are employed to include citations in their content, grounding the content in verifiable sources. Despite such developments, manually assessing how well a citation supports the associated statement remains a major challenge. Previous studies tackle this challenge by leveraging faithfulness metrics to estimate citation support automatically. However, they limit this citation support estimation to a binary classification scenario, neglecting fine-grained citation support in practical scenarios. To investigate the effectiveness of faithfulness metrics in fine-grained scenarios, we propose a comparative evaluation framework that assesses the metric effectiveness in distinguishing citations between three-category support levels: full, partial, and no support. Our framework employs correlation analysis, classification evaluation, and retrieval evaluation to measure the alignment between metric scores and human judgments comprehensively. Our results indicate no single metric consistently excels across all evaluations, highlighting the complexity of accurately evaluating fine-grained support levels. Particularly, we find that the best-performing metrics struggle to distinguish partial support from full or no support. Based on these findings, we provide practical recommendations for developing more effective metrics.