Event type extraction inference

[AndDoIt]

I have finished the training of Entity type extraction task, and I have a file with raw text and no event label, I wondered how to get the entity type inference?

[boy56]

Entity type extraction? Is it like Argument Extraction?
Well, In the inference process, you can do
(1) Use a classification model(ours or other methods like bert) to get the event types of the raw text
(2) Take the event type, role type(related to the event type) and the raw text as the input, and the model will generate the arguments(or entities) of this role type.

[AndDoIt]

Thanks very much for your reply!
I just finished the model training of first module of Event Type Dection, and now I want to use your model that I have trained to get the event types of the raw text. While the other two tasks have the inference bash script, for my first task, what kind of format that dataset shoule be converted? And just use run_seq2seq.py with argument --do_predict?

[boy56]

Thanks very much for your reply! I just finished the model training of first module of Event Type Dection, and now I want to use your model that I have trained to get the event types of the raw text. While the other two tasks have the inference bash script, for my first task, what kind of format that dataset shoule be converted? And just use run_seq2seq.py with argument --do_predict?

I recommend you that:
(1) use the data_convert/convert_text_to_target.py to convert the data format.
(2) train the argument extraction model use the gold event type (bash run_seq2seq_span.bash XXX).
(3) use the argument extraction model to do this task: event type role type raw text -> roles. (run_arg_predict.bash will be useful).

[AndDoIt]

Okay, I see. Thanks for your kindly reply!