Large Language Model-Driven Knowledge Graph Construction for Photoresist Materials

Vast knowledge dispersed across the unstructured literature remains an indispensable complement to photoresist development. However, the intricate relationships among formulation, processing, and performance are not only scattered throughout publications but also complicated by semantic heterogeneity and contextual ambiguity, hindering knowledge management that extends beyond simple retrieval. To address these issues, we employed large language models (LLMs) to construct a photoresist material knowledge graph (KG) covering synthesis, properties, and applications. From nearly 12,000 full-text documents, we extracted 320,000 entities, 178,000 relationships, and 1,172,000 attribute entries. By integrating this KG with an LLM via a dual-path retrieval-augmented generation (RAG) strategy, we achieved 95.33% accuracy on a domain-specific question-answering task, substantially outperforming the standalone LLM (76.67%). This work offers a reference framework for knowledge management and intelligent reasoning in the photoresist field.