lavis-nlp / spert

PyTorch code for SpERT: Span-based Entity and Relation Transformer
MIT License
692 stars 148 forks source link

Parts of entities are recognised separately #58

Closed maiiabocharova closed 3 years ago

maiiabocharova commented 3 years ago

Hello, can you please help me with advice how can I solve this issue: trained SpERT on my custom dataset and am getting most of the entities right, but there is this strange repetition with parts of the entities. Maybe there is some way to solve it. (Didn't have such an issue with huggingface transformers for token classification) {"type": "3.2_mixtures/components/concentration/value", "start": 82, "end": 86}, {"type": "3.2_mixtures/components/concentration/value", "start": 82, "end": 87}, {"type": "3.2_mixtures/components/concentration/value", "start": 82, "end": 88}, {"type": "3.2_mixtures/components/concentration/value", "start": 83, "end": 87}, {"type": "3.2_mixtures/components/concentration/value", "start": 83, "end": 88}, {"type": "3.2_mixtures/components/concentration/value", "start": 84, "end": 87}, {"type": "3.2_mixtures/components/concentration/value", "start": 84, "end": 88}, {"type": "3.2_mixtures/components/concentration/value", "start": 85, "end": 87}, {"type": "3.2_mixtures/components/concentration/value", "start": 85, "end": 88}, {"type": "3.2_mixtures/components/concentration/value", "start": 86, "end": 88},

markus-eberts commented 3 years ago

Hi, are there any control characters in your input sentence (e.g. \n)? Could you please send me the corresponding sentence? Also, how many negative entity samples do you use ('neg_entity_count' in the configuration file)?

markus-eberts commented 3 years ago

We had a conversation by email. Please leave a comment if the issue is still relevant.

maiiabocharova commented 3 years ago

Hi, sorry didn't answer you, got ill (had covid), but changing the number of negative samples helped!

Best regards, Maiia.

вт, 19 окт. 2021 г. в 16:27, Markus Eberts @.***>:

We had a conversation by email. Please leave a comment if the issue is still relevant.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/lavis-nlp/spert/issues/58#issuecomment-946722871, or unsubscribe https://github.com/notifications/unsubscribe-auth/AQ7UPWU2KYNK6CRKSNHEZFDUHVW2RANCNFSM5E2WEIFA . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub.