Improving Information Extraction with Learned Queries

Authors/institution/venue: Omar Sharif, Soroush Vosoughi and Nikhil Singh; affiliations not specified on the abstract page; arXiv preprint submitted 31 August. Problem: clinical extraction quality depends heavily on query design. Contribution: learned and feedback-optimised queries across four clinical benchmarks and five LLMs; authors report an 18.6 F1 gain and release a 12,820-question dataset. Significance: smaller models may become viable for narrow extraction. Limitations: preprint, benchmark generalisation, privacy and clinical validation. Paper