Data-Driven Discovery of Lithium–Sulfur Catalysts via Two-Stage Large Language Model Fine-Tuning and Mechanistic Validation

Status: Published in Materials Today Energy 59 (2026), Article 102371.

Role: First author and workflow leader.

Contribution: Developed the overall literature-driven screening framework and coordinated scientific-document processing, domain-specific dataset construction, two-stage model fine-tuning, multidimensional scoring, model comparison, external consistency evaluation, mechanistic analysis, manuscript preparation, and revision.

This study develops a literature-driven framework for evaluating catalysts reported in lithium–sulfur battery research. The workflow integrates paper-aware scientific information extraction, domain-specific dataset construction, two-stage language-model fine-tuning, multidimensional catalyst scoring, external consistency evaluation, and mechanistic validation.

The project investigates whether lightweight domain-adapted language models can support scalable and interpretable catalyst assessment across heterogeneous scientific literature while reducing the computational cost of domain-specific screening.

Recommended citation: Liu, H.; Li, S.; Huang, Y.; Zhao, Z.; Wu, G.; Wang, N.; Sun, Z.; Liu, C.; Yan, H.; Wu, Z.; Wang, Z.; Liu, E.; Shan, Z.; Shi, C.; He, F.; He, C.; Zhao, N.; Tang, Y.; Hu, Y.; Chen, B. (2026). "Data-Driven Discovery of Lithium–Sulfur Catalysts via Two-Stage Large Language Model Fine-Tuning and Mechanistic Validation." Materials Today Energy 59, 102371.
Download Paper