what should I do if the sentence is longer more？Do you have any suggests，please？

WindChimeRan / pytorch_multi_head_selection_re

BERT + reproduce "Joint entity recognition and relation extraction as a multi-head selection problem" for Chinese and English IE

138 stars 31 forks source link

what should I do if the sentence is longer more？Do you have any suggests，please？ #16

Closed moyan007 closed 4 years ago

moyan007 commented 4 years ago

Thanks for your sharing firstly! And I am working on the Joint entity recognition and relation extraction over a long text which the average length is about 1000 tokens. Could you please give me some advices for that? I have tried to adjust the length of config json file, but the performance is very pool. Like this: Triplets-> p: 0.0000, r: 0.0000, f: 0.0000 ||NER->p: 1.0000, r: 0.0000, f: 0.0000

WindChimeRan commented 4 years ago

No, you should not use MHS for long sentences, because the space complexity is O(relation x length x length). You can try my new method: Seq2UMTree.