search

dwadden / dygiepp

Span-based system for named entity, relation, and event extraction.
MIT License
573 stars 120 forks source link

Genia dataset for Event Annotations #82

Closed Toan-it-mta closed 2 years ago

Toan-it-mta commented 2 years ago

I see in "https://github.com/dwadden/dygiepp/blob/master/doc/data.md#preprocesing-details-for-existing-datasets" mentioned dataset Genia can use to Event Annotations . I would like to ask, does your code currently support Genia data preprocessing for the Event Annotations task? If yes, how to do?

Thanks

dwadden commented 2 years ago

The code does not support GENIA for events, if I recall. I'd welcome a pull request, though! One challenge with GENIA is that event arguments can themselves be other events, and it's not totally intuitive how to cast this as a span-level task. One possible way to do this: Suppose there's an event A, triggered by span t_a, and event B, triggered by t_b. If event B is an argument for event A, then use the span t_b to represent event B as the argument for A. This would be represented in DyGIE as a directed edge from t_a to t_b, the same way that DyGIE represents relations as directed edges between pairs of spans. Let me know if this makes sense.

Toan-it-mta commented 2 years ago

Thanks for the answer and suggestion from you. I am new in this field. I will try your suggestion and ask for permission to close the issue.

dwadden commented 2 years ago

Sure, sounds good. If you decide to attempt a PR, let me know and I'm happy to try and help if I can.