LLMJudge: LLMs for Relevance Judgments

TL;DR

LLMJudge tackles scalable relevance judgments for IR by leveraging LLMs to generate labels, mitigating the cost of human annotations. The paper designs a four-point relevance scoring task grounded in the TREC-DL 2023 dataset, compares open-source and closed-source LLMs, and analyzes prompts, biases, and data leakage, with data and tooling released for reproducible research. Evaluation combines Cohen's $\kappa$ for label agreement and Kendall's $\tau$ for system ordering, based on 39 submissions from seven groups, revealing meaningful variation in exact judgments but stable system rankings. Overall, the work provides a public benchmark and methodology for automatic relevance judgment in IR and search, enabling broader, data-efficient evaluation of IR systems.

Abstract

The LLMJudge challenge is organized as part of the LLM4Eval workshop at SIGIR 2024. Test collections are essential for evaluating information retrieval (IR) systems. The evaluation and tuning of a search system is largely based on relevance labels, which indicate whether a document is useful for a specific search and user. However, collecting relevance judgments on a large scale is costly and resource-intensive. Consequently, typical experiments rely on third-party labelers who may not always produce accurate annotations. The LLMJudge challenge aims to explore an alternative approach by using LLMs to generate relevance judgments. Recent studies have shown that LLMs can generate reliable relevance judgments for search systems. However, it remains unclear which LLMs can match the accuracy of human labelers, which prompts are most effective, how fine-tuned open-source LLMs compare to closed-source LLMs like GPT-4, whether there are biases in synthetically generated data, and if data leakage affects the quality of generated labels. This challenge will investigate these questions, and the collected data will be released as a package to support automatic relevance judgment research in information retrieval and search.

Figures (1)

• Figure 1: Scatter plot of Cohen's $\kappa$ and Kendall's $\tau$ for submitted labelers