Accelerating materials language processing with large language models

This paper introduces the use of large language models (LLMs), specifically generative pre-trained transformers (GPT), to enhance materials language processing (MLP). The authors address challenges associated with complex model architectures and extensive fine-tuning in traditional MLP approaches by employing strategic prompt engineering with GPT models. Their findings demonstrate high performance in text classification, named entity recognition (NER), and extractive question answering (QA) using limited datasets, even with zero-shot or few-shot learning. The GPT-based approach is shown to be effective across various materials classes and can assist materials scientists in knowledge-intensive MLP tasks, even without specialized expertise. Furthermore, the authors highlight the potential of GPT models to reduce researcher workload by providing initial labeling sets and validating human annotations.

Jaewoong Choi, Byungju Lee