dkorenci / pan-clef-2024-oppositional Goto Github PK

This repository contains code with utilities, baselines, instructions, and guidelines for the
PAN CLEF 2024 shared task "Oppositional thinking analysis: Conspiracy vs critical narratives"
https://pan.webis.de/clef24/pan24-web/oppositional-thinking-analysis.html
The code is licensed under the APACHE 2.0 license, see LICENSE file.
The exception is the span_f1_metric.py file, which is licensed under the GPL license, see the file for details.

This document contains the high-level overview of the code,
and the instructions to setup the environment for the project.

For the details of the data, see README-DATA,
for the conceptual overview of the task, and the guidelines and possible approaches for the participants,
see the slides in pan2024_oppositional_overview_guidelines.pdf

For the details on the evaluation metrics and procedure, see the README-EVALUATION file.

IMPORTANT: keep up with the updates on the official repository,
as the code and especially the README files might be improved and updated.


# 1. High-level Overview

The code is organized into three packages:

- data_tools
loading and preprocessing of the data

- classif_experim
baselines for the classification task, and the supporting functionality

- sequence_labeling
baselines for the sequence labeling task, and the supporting functionality

- getting started
First, create and modify the local variants of .gitignore and setting.py (from templates)
only templates are git-tracked, to facilitate the local setup
Second, setup the environment (see below). Third, run the experiments (see below).


# 2. Setup of the environment using conda

conda create --name env-name python=3.X # 3.10 is recommended
conda activate env-name
# from the root of the repository
./generate_requirements.sh  # mind the additional dependencies, see the code
pip install -r requirements.txt

If you will be relying on spacy docs as input format while working
with the seq. labeling baseline (this is the default so you probably will), do:
python -m spacy download en_core_web_sm
python -m spacy download es_core_news_sm


# 3. Running the experiments
Rename setting_template.py to settings.py and fill in the paths to the dataset files.

python entrypoint for running the classification experiment (subtask 1) is:
classif_experiment_runner.run_classif_experiments

python entrypoint for running the seq. labeling experiment () is:
seqlabel_experiment_runner.run_seqlab_experiments
CLI entrypoint is the seqlabel_experiment_runner.main,
you can run the module as in: run_seqlabel.sh.template, create the local .sh copy first