This container adds argument mining to the ADD-up framwork. In detail: We integrate the backend as provided by Chernodub et al. (2019): TARGER: Neural Argument Mining at Your Fingertips. The backend provides an API to detect arguments. Arguments are printed in semantic role labeling BIO notation, for an example see e.g. Habernal/Gurevych 2017: Argument Mining in User-Generated Web Discourse.

The reference of TARGER is:

Chernodub, Artem and Oliynyk, Oleksiy and Heidenreich, Philipp and Bondarenko, Alexander and Hagen, Matthias and Biemann, Chris and Panchenko, Alexander (2019): TARGER: Neural Argument Mining at Your Fingertips, Proceedings of the 57th Annual Meeting of the Association of Computational Linguistics (ACL 2019), Florence, Italy

More information can be found here: http://github.com/achernodub/targer
There is also an online demo available: http://ltdemos.informatik.uni-hamburg.de/targer/
API reference: http://ltdemos.informatik.uni-hamburg.de/arg-api#/

Build container

  1. The model(s) must be downloaded separately, the container does not include the models. You can download the models at http://ltdata1.informatik.uni-hamburg.de/targer/. Please put the models in the directory 00_custom/targer_models.

  2. Run (regular) docker-compose up command (takes a while to compile)

  3. Access API via curl: curl -X POST “http://localhost:8001/classifyCombo” -H “accept: application/json” -H “Content-Type: text/plain” -d “Quebecan independence is justified. In the special episode in Japan, his system is restored by a doctor who wishes to use his independence for her selfish reasons.”

  4. To use Targer argument mining, please set the environment variable RUN_TARGER to 1. Please also define the model to use, e.g. TARGER_MODEL=Combo.h5

Main Contributor

Valentin Gold