Good idea to improve the term recognition! But using embeddings and the neural stuff might be too complex and overkill. Simpler, older methods like the Lesk family of algorithms and derivatives (https://en.wikipedia.org/wiki/Lesk_algorithm) might be easier to implement and could actually make use of (and thus promote the creation of) definitions and context sentences available in the termbase and linked ontologies etc.

Good idea to improve the term recognition! But using embeddings and the neural stuff might be too complex and overkill. Simpler, older methods like the Lesk family of algorithms and derivatives (https://en.wikipedia.org/wiki/Lesk_algorithm) might be easier to implement and could actually make use of (and thus promote the creation of) definitions and context sentences available in the termbase and linked ontologies etc.

Thank you! Your suggestion sounds good.