This release of ASMR-E (Automatic Software Mention Recognition and Extraction) introduces significant improvements in software mention extraction from academic texts. The system consists of two distinct pipelines, each designed for different stages of extraction and verification, ensuring both broad coverage and high precision in software mention identification.
The system leverages LLM-based processing, semantic filtering, and keyword-based heuristics to accurately identify software names, versions, and associated metadata from structured and unstructured textual sources.
This release includes optimizations to both extraction pipelines, enhancements in entity verification, and more efficient handling of JSON outputs for interoperability with external ontologies.