search

Nealcly / templateNER

Source code for template-based NER
208 stars 39 forks source link

Seq2SeqModel predicts one entity at a time #13

Closed savasy closed 2 years ago

savasy commented 2 years ago

Hi, Seq2SeqModel.predict function predicts one entity at a time. e.g.

predict("Tesla, IBM, and Amazon are the good tech companies") -> "Tesla is an Organization" What about extracting IBM and Amazon as well at the same time!

Nealcly commented 2 years ago

During inference, we do not generate the prediction directly. We consider the model as the scoring function. Given a sentence, we first enumerate all possible text spans in the input sentence as named entity candidates, and then classify them into entities or non-entities based on BART scores on templates.

savasy commented 2 years ago

Thank you @Nealcly My problem is that I do not calculate the detailed performance (F1, Prec etc.) for different Label sets other than PER LOC ORG. I managed to train model with 9 labels and get a good accuracy btw. When I use your Inference.py code to run the model trained with 9 labels, I got some errors. Maybe I share it with a different issue thread