Support Evaluation for the TREC 2024 RAG Track: Comparing Human versus LLM Judges

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Thakur, Nandan, Pradeep, Ronak, Upadhyay, Shivani, Campos, Daniel, Craswell, Nick, Lin, Jimmy
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866917993011740672
author Thakur, Nandan
Pradeep, Ronak
Upadhyay, Shivani
Campos, Daniel
Craswell, Nick
Lin, Jimmy
author_facet Thakur, Nandan
Pradeep, Ronak
Upadhyay, Shivani
Campos, Daniel
Craswell, Nick
Lin, Jimmy
contents Retrieval-augmented generation (RAG) enables large language models (LLMs) to generate answers with citations from source documents containing "ground truth", thereby reducing system hallucinations. A crucial factor in RAG evaluation is "support", whether the information in the cited documents supports the answer. To this end, we conducted a large-scale comparative study of 45 participant submissions on 36 topics to the TREC 2024 RAG Track, comparing an automatic LLM judge (GPT-4o) against human judges for support assessment. We considered two conditions: (1) fully manual assessments from scratch and (2) manual assessments with post-editing of LLM predictions. Our results indicate that for 56% of the manual from-scratch assessments, human and GPT-4o predictions match perfectly (on a three-level scale), increasing to 72% in the manual with post-editing condition. Furthermore, by carefully analyzing the disagreements in an unbiased study, we found that an independent human judge correlates better with GPT-4o than a human judge, suggesting that LLM judges can be a reliable alternative for support assessment. To conclude, we provide a qualitative analysis of human and GPT-4o errors to help guide future iterations of support assessment.
format Preprint
id arxiv_https___arxiv_org_abs_2504_15205
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Support Evaluation for the TREC 2024 RAG Track: Comparing Human versus LLM Judges
Thakur, Nandan
Pradeep, Ronak
Upadhyay, Shivani
Campos, Daniel
Craswell, Nick
Lin, Jimmy
Computation and Language
Artificial Intelligence
Information Retrieval
Retrieval-augmented generation (RAG) enables large language models (LLMs) to generate answers with citations from source documents containing "ground truth", thereby reducing system hallucinations. A crucial factor in RAG evaluation is "support", whether the information in the cited documents supports the answer. To this end, we conducted a large-scale comparative study of 45 participant submissions on 36 topics to the TREC 2024 RAG Track, comparing an automatic LLM judge (GPT-4o) against human judges for support assessment. We considered two conditions: (1) fully manual assessments from scratch and (2) manual assessments with post-editing of LLM predictions. Our results indicate that for 56% of the manual from-scratch assessments, human and GPT-4o predictions match perfectly (on a three-level scale), increasing to 72% in the manual with post-editing condition. Furthermore, by carefully analyzing the disagreements in an unbiased study, we found that an independent human judge correlates better with GPT-4o than a human judge, suggesting that LLM judges can be a reliable alternative for support assessment. To conclude, we provide a qualitative analysis of human and GPT-4o errors to help guide future iterations of support assessment.
title Support Evaluation for the TREC 2024 RAG Track: Comparing Human versus LLM Judges
topic Computation and Language
Artificial Intelligence
Information Retrieval
url https://arxiv.org/abs/2504.15205