Retrieving Argument Graphs Using Vision Transformers

This is the repository for our paper "Retrieving Argument Graphs Using Vision Transformers". It contains our code

• to visualize argumentation graphs using our visualizations: Treemaps, Logical and Space-Reclaiming Icicle Plots (vis directory),
• for pre-training and fine-tuning our vision transformers (training directory),
• for generating a training dataset to fine-tune a GPT-4o model (oai_finetuning directory),
• to evaluate the various models (eval directory).

A more detailed explanation of our scripts can be found in the README.md within those directories.

Data

A corpus of publicly available argumentation graphs is available on GitHub. For our evaluation, we used the microtexts dataset, together with the requests found here.

Installation

The project's dependencies are managed by uv. After installing uv, create a virtual environment for the project using uv sync. After that, the various scripts can be executed using uv run python <script>.