Python notebook of Princeton GraphMERT Paper – a better knowledge graph (github.com)

A community-contributed Python notebook reimplementation of Princeton’s GraphMERT paper — “GraphMERT: Efficient and Scalable Distillation of Reliable Knowledge Graphs from Unstructured Data” (Belova et al., arXiv:2510.09580, 10 Oct 2025) has been published as a simple test example. The notebook provides a hands-on, runnable example that demonstrates the paper’s core idea: distilling high-quality, reliable knowledge graphs from raw text in an efficient and scalable way. All rights belong to the original authors; the notebook is positioned as a reproducibility and onboarding tool rather than a production system. Why this matters: GraphMERT targets two persistent pain points in KG construction — noise from unstructured sources and the compute cost of large-scale graph extraction. By providing a tested implementation, the notebook lowers the barrier for researchers and engineers to reproduce results, inspect intermediate outputs (extraction, filtering, scoring), and adapt the pipeline to new corpora or downstream tasks such as QA, retrieval, and knowledge-grounded generation. Technically the notebook offers a compact demonstration of the paper’s distillation pipeline, with code for preprocessing, candidate extraction, reliability scoring/filtering, and basic evaluation metrics — a practical starting point for benchmarking, extending the method, or integrating GraphMERT components into ML/IR workflows.