lubianat / ann

A repository for brainstorming and prototyping ideas related to the eLifeSprint project Annotate them all (https://sprint.elifesciences.org/annotate-them-all/)
Apache License 2.0
14 stars 0 forks source link

Increase accuracy of Wikidata API search #23

Closed arupgsh closed 4 years ago

arupgsh commented 4 years ago

Is there any way to classify the entity based on the category and context.

Eg.
Apple Inc > Company
Apple > Fruit
webersab commented 4 years ago

I read the paper and it seems like sci spacy does not use contextualized word embeddings (the model they usde is described here: https://arxiv.org/pdf/1603.01360.pdf). There is some code for detecing and normalizing biomedical entities using contextualized word embeddings here: https://bern.korea.ac.kr/ We could try and see if this is better at detecting the entities than sci spacy. We would still have to link the entites to Wikidata, but maybe we could use that in addition to sci spacy.