ScienceBeam GROBID bioRxiv

Overview

Note: This has been largely replaced by ScienceBeam Parser.

A variation of the GROBID image, that includes models trained on bioRxiv. It builds on top of changes in the following GROBID fork and sciencebeam-trainer-delft for the DL models.

Run Image

There are multiple image tags of elifesciences/sciencebeam-grobid-biorxiv, with the following tag suffixes:

tag suffix	description
wapiti	traditional, non-DL model
dl-no-word-embeddings	DL model, not using any word embeddings
dl-no-word-embeddings-wapiti-citation	Same as dl-no-word-embeddings, but the citation model is using wapiti.
dl-glove-6B-50d	DL model, using glove.6B 50d word embeddings. This improves the accuracy over not using any word embeddings while still keeping the image size reasonable. (these models are currently not updated)

All of the models are trained using the same bioRxiv dataset.

The latest tag will be set to the latest dl-glove-6B-50d image.

docker pull elifesciences/sciencebeam-grobid-biorxiv
docker run -t --rm --init -p 8080:8070 -p 8081:8071 \
  elifesciences/sciencebeam-grobid-biorxiv

(for deployments it is recommended to use a specific tag)

See also GROBID and containers