Back to Search Start Over

Improving accuracy of GPT-3/4 results on biomedical data using a retrieval-augmented language model

Authors :
Soong, David
Sridhar, Sriram
Si, Han
Wagner, Jan-Samuel
Sá, Ana Caroline Costa
Yu, Christina Y
Karagoz, Kubra
Guan, Meijian
Hamadeh, Hisham
Higgs, Brandon W
Source :
PLOS Digit Health, 3(8) , 2024
Publication Year :
2023

Abstract

Large language models (LLMs) have made significant advancements in natural language processing (NLP). Broad corpora capture diverse patterns but can introduce irrelevance, while focused corpora enhance reliability by reducing misleading information. Training LLMs on focused corpora poses computational challenges. An alternative approach is to use a retrieval-augmentation (RetA) method tested in a specific domain. To evaluate LLM performance, OpenAI's GPT-3, GPT-4, Bing's Prometheus, and a custom RetA model were compared using 19 questions on diffuse large B-cell lymphoma (DLBCL) disease. Eight independent reviewers assessed responses based on accuracy, relevance, and readability (rated 1-3). The RetA model performed best in accuracy (12/19 3-point scores, total=47) and relevance (13/19, 50), followed by GPT-4 (8/19, 43; 11/19, 49). GPT-4 received the highest readability scores (17/19, 55), followed by GPT-3 (15/19, 53) and the RetA model (11/19, 47). Prometheus underperformed in accuracy (34), relevance (32), and readability (38). Both GPT-3.5 and GPT-4 had more hallucinations in all 19 responses compared to the RetA model and Prometheus. Hallucinations were mostly associated with non-existent references or fabricated efficacy data. These findings suggest that RetA models, supplemented with domain-specific corpora, may outperform general-purpose LLMs in accuracy and relevance within specific domains. However, this evaluation was limited to specific questions and metrics and may not capture challenges in semantic search and other NLP tasks. Further research will explore different LLM architectures, RetA methodologies, and evaluation methods to assess strengths and limitations more comprehensively.

Details

Database :
arXiv
Journal :
PLOS Digit Health, 3(8) , 2024
Publication Type :
Report
Accession number :
edsarx.2305.17116
Document Type :
Working Paper
Full Text :
https://doi.org/10.1371/journal.pdig.0000568