NER - caught in the middle of an entity, starting with "I-" instead of "B-" #1699
Open
Description
Hi,
This relates to #1657
Model ner_mult_long_demo
There are several text inputs that return tags/labels that start with "I-" rather than "B-".
Example:
This is a public service announcement.
This returns:
[[["This","is","a","public","service","announcement","."],["O","O","O","O","I-EVENT_NAME","I-EVENT_NAME","O"]]]
See how there's no B-EVENT_NAME
.
docker-compose.yml:
ner:
container_name: ner
image: deeppavlov/deeppavlov
environment:
- CONFIG=ner_mult_long_demo
restart: always
volumes:
- ./deeppavlov/ner_mult_long_demo.json:/usr/local/lib/python3.10/site-packages/deeppavlov/configs/classifiers/ner_mult_long_demo.json
- ./deeppavlov/sentence_delimiter.py:/usr/local/lib/python3.10/site-packages/deeppavlov/models/tokenizers/sentence_delimiter.py
- ./deeppavlov/pysbd.txt:/usr/local/lib/python3.10/site-packages/deeppavlov/requirements/pysbd.txt
- ./deeppavlov/registry.json:/usr/local/lib/python3.10/site-packages/deeppavlov/core/common/registry.json
- ./deeppavlov/requirements_registry.json:/usr/local/lib/python3.10/site-packages/deeppavlov/core/common/requirements_registry.json
- ./data:/root/.deeppavlov
- ./venv:/venv
entrypoint:
- /bin/sh
- -c
- |
/usr/local/bin/python3.10 -m pip install pysbd==0.3.4
python -m deeppavlov riseapi ner_mult_long_demo -p 5000 -d
This doesn't happen in the demo of the website, nor if I use the model ner_demo_mdeberta_address