Document Attribution: Examining Citation Relationships using Large Language Models

TL;DR

This paper investigates methods for improving document attribution in large language models, proposing zero-shot entailment techniques and analyzing attention mechanisms to enhance citation reliability and interpretability.

Contribution

It introduces a zero-shot entailment approach for attribution and examines the impact of attention layers on attribution performance in LLMs.

Findings

Zero-shot entailment improves attribution accuracy on benchmark datasets.

Attention mechanisms influence attribution performance across different layers.

Flan-ul2 outperforms baselines in citation correctness metrics.

Abstract

As Large Language Models (LLMs) are increasingly applied to document-based tasks - such as document summarization, question answering, and information extraction - where user requirements focus on retrieving information from provided documents rather than relying on the model's parametric knowledge, ensuring the trustworthiness and interpretability of these systems has become a critical concern. A central approach to addressing this challenge is attribution, which involves tracing the generated outputs back to their source documents. However, since LLMs can produce inaccurate or imprecise responses, it is crucial to assess the reliability of these citations. To tackle this, our work proposes two techniques. (1) A zero-shot approach that frames attribution as a straightforward textual entailment task. Our method using flan-ul2 demonstrates an improvement of 0.27% and 2.4% over the best baseline of ID and OOD sets of AttributionBench, respectively. (2) We also explore the role of the attention mechanism in enhancing the attribution process. Using a smaller LLM, flan-t5-small, the F1 scores outperform the baseline across almost all layers except layer 4 and layers 8 through 11.