Reconstructing knowledge on synthesis routes and properties from inorganic science literature is crucial yet challenging, particularly in maintaining completeness and logical consistency. Here, the authors develop a generalized method based on GPT-4 to fine-tune LLMs, achieving high precision in material synthesis extraction and demonstrating broad applicability across domains, ultimately constructing a comprehensive knowledge graph for materials science optimization.
- Shuyuan Li
- Shihao Wei
- Shaorui Sun
