MAMKit

MAMKit is an open-source, publicly available PyTorch toolkit designed to access and develop datasets, models, and benchmarks for Multimodal Argument Mining (MAM). It provides a flexible interface for accessing and integrating datasets, models, and preprocessing strategies through composition or custom definition. MAMKit is designed to be extendible, ensure replicability, and provide a shared interface as a common foundation for experimentation in the field.

Currently, MAMKit offers 4 datasets, 4 tasks and 6 distinct model architectures, along with audio and text processing capabilities, organized in 5 main components.

Structure

The toolkit is organized into five main components: configs, data, models, modules and utility. In addition to that, the toolkit provides a demos directory for running all the experiments presented in the paper. The figure below illustrates the toolkit’s structure.

Install

pip

pip install mamkit

git

git clone https://github.com/lt-nlp-lab-unibo/mamkit

Contribute

Feel free to submit a pull request! We welcome new datasets, models, and any other contribution that can improve the toolkit!

MAMKit is meant to be a community project :)

Contact

Don’t hesitate to contact: - Eleonora Mancini - Federico Ruggeri

for questions/doubts/issues!

Citing

If you use MAMKit in your research, please cite the following paper:

@inproceedings{TBAmamkit,
  title={MAMKit: A Comprehensive Multimodal Argument Mining Toolkit},
  author={TBA},
  booktitle={TBA},
  year={TBA}
}