KEVLAR (Kessler EuroVoc Laws and Acts Repository) is a set of tools and datasets related to EU laws. It includes an archive of EU legal documents in 24 languages, covering from 1951 to 2022 and downloaded from EUR-Lex, the software to train a model able to classify documents with respect to the EuroVox taxonomy, a package to run the previous tool as an API, and a web demo.
Authors: Lorenzo Bocchi and Alessio Palmero Aprosio.
Some pre-processing tools are available in different Github projects. In particular, ScrapeLex can be used to download and collect legal documents from EUR-Lex, and AutoEuroVoc contains all the code to train and test the models.
The documents can also be downloaded from the KEVLAR OneDrive repository. Both the models and the test data are available on the Digital Humanities KEVLAR repository.
In this project one can find the Dockerfile that can be used to build an out-of-the-box working environment containing the classifier (REST API), and the web demo.
To execute the software using the provided Dockerfile, clone/download the repository and build the image using the command:
docker build --build-arg="LANGS=XXX" -t kevlar .
where XXX is a space-separated list of languages (for instance, it en).
The command can be executed without the --build-arg parameter. In this case, all the 24 models are downloaded and included (more than 50GB in total).
When the build is complete, run the command:
docker run -p 8080:80 kevlar
where 8080 can be replaced with any other port number, depending of the needs.
In both commands, kevlar can be replaced by a custom name.