Shared Task on Information Extraction for Machine Learning

Reproducing machine learning research requires structured access to the models, datasets, methods, and relations described in scientific publications. IE4ML builds on GSAP-ERE, an extension of GSAP-NER, and turns its schema into a community benchmark for producing machine-actionable descriptions suitable for scholarly knowledge graphs and Web-scale discovery.

The task uses 10 entity and 18 relation types and comprises two subtasks: 1. IE4ML Entity Extraction (span extraction and classification); 2. IE4ML Relation Extraction (given gold entity spans). Training uses 80 publicly released GSAP-ERE publications; validation and test sets each contain 20 newly annotated papers released during the evaluation phases.