Awesome

grobid-astro

The goals of this GROBID module are:

to recognize in textual documents and PDF any mentions of astronomical objects,
to disambiguate the mention with respect to the astronomical knowledge base SIMBAD.

As the other GROBID models, the module relies only on machine learning and uses linear CRF.

Example: Astronomical object recognition in a text segment with link to Simbad entry

Install, build, run

Building grobid-astro and JDK 1.8 (gradle wrapper is shipped with the code).

First install the latest development version of GROBID as explained by the documentation.

Copy the module astro as sibling sub-project to grobid-core, grobid-trainer, etc.:

cp -r grobid-astro grobid/

cd grobid-astro/

Copy the existing trained model in the standard grobid-home path, type the command under grobid/grobid-atro path:

./gradlew copyModels

Try compiling everything with:

./gradlew clean build

Run some test:

cd PATH-TO-GROBID/grobid/grobid-astro

./gradlew test

The models have to be trained before running the tests! - See bellow for training an astro model.

Start the service

./gradlew appRun

Demo/console web app is then accessible at http://localhost:8080. Remember to close it by pressing a button and not via ctrl+c or the java process won't be killed.

Using curl POST/GET requests:

curl -X POST -d "text=Look at GRB 020819, on the right." localhost:8080/service/processAstroText

curl -GET --data-urlencode "text=Look at Andromeda below the North Star." localhost:8080/service/processAstroText

Training and evaluation

Training only

For training the astro model with all the available training data:

> cd PATH-TO-GROBID/grobid/grobid-astro

> ./gradlew train_astro

The training data must be under grobid-astro/resources/dataset/astro/corpus

Evaluating only

For evaluating under the labeled data under grobid-astro/resources/dataset/astro/evaluation, use the command:

>  ./gradlew eval_astro

Training and evaluating with automatic corpus split

The following commands will split automatically and randomly the available annotated data (under resources/dataset/astro/corpus/) into a training set and an evaluation set, train a model based on the first set and launch an evaluation based on the second set.

>  ./gradlew eval_astro_split

In this mode, by default, 80% of the available data is used for training and the remaining for evaluation. This ratio can be changed by editing the corresponding exec profile in the pom.xml file.

Training data

...

Generation of training data

For generating training data in TEI, based on the current model, from a list of text or PDF files in a input repository, use the following command:

> java -Xmx4G -jar build/libs/grobid-astro-0.6.2-SNAPSHOT-onejar.jar -gH ../grobid-home -dIn ~/test_astro/ -dOut ~/test_astro/out/ -exe createTraining

(change the grobid-astro version as required!)

License

GROBID is distributed under Apache 2.0 license.