SciLitMiner: An Intelligent System for Scientific Literature Mining and Knowledge Discovery

Recent advances in data mining have enabled automation in literature‐based discovery (LBD), allowing synergistic evaluation of experimental findings reported in scientific publications. However, existing tools and digital libraries fall short in generating relevant literature collections and evaluating them for highly specific questions. This article presents SciLitMiner, an intelligent system to address this gap. SciLitMiner enables federated ingestion of literature from digital libraries; applies advanced retrieval techniques, including dataset‐aware retrieval from visual elements, to identify relevant studies; and leverages retrieval‐augmented generation (RAG) tailored to domain‐specific knowledge reasoning. The system is applied in materials science to study the creep behavior of γ‐TiAl alloys, revealing the intricate interplay between material, process, microstructure, and creep rate, represented through knowledge graphs. Two domain experts rate responses from the knowledge reasoning workflow with OpenAI large language models (LLMs) as the backbone above “good” (>3 on a 5‐point Likert scale) in over 90% across qualitative criteria, indicating strong performance. In a case study, the workflow also outperforms gpt‐4.5‐turbo with web search and other leading tools in reliability. A second case study benchmarks open‐source LLMs as drop‐in replacements for proprietary models, demonstrating comparable‐to‐superior performance. The system's flexibility enables its use in automated LBD across diverse research domains.