KnowMat: Transforming Unstructured Material Science Literature into Structured Knowledge

The rapid expansion of scientific literature in materials science presents challenges for efficiently extracting and analyzing experimental data. To address this, we introduce KnowMat, an accessible pipeline that transforms unstructured texts into structured, machine-readable datasets. Leveraging lightweight open-source Large Language Models (LLMs), such as Llama 3.1 (8B) and Llama 3.2 (3B) through the Ollama platform, KnowMat automatically extracts key materials information including composition, processing conditions, characterization methods, and properties. Implemented via an intuitive Flask-based web interface, users can easily upload documents, manage extraction tasks, and save structured outputs directly to CSV files, facilitating database creation and integration into machine learning workflows. Evaluation on real-world materials science papers demonstrates KnowMat’s accuracy, efficiency, and usability on consumer-grade hardware, significantly reducing barriers to data-driven materials research.