Greek Article Classifier

The Greek Article Classifier is a Python project that classifies Greek news articles using large language models (LLMs) with zero-shot prompts. In addition to its main classification functionality, the repository includes separate evaluation and baseline modules.

Table of Contents

• Greek Article Classifier
  • Table of Contents
  • Overview
  • Installation
  • Command Line Arguments
  • Usage
    • Main Program
    • Evaluation
    • BERT Baseline
  • License

Overview

This project uses LLMs to perform zero-shot classification of Greek news articles. The main entry point (main.py) provides various command-line options to customize behavior, such as specifying a custom client, saving results, and adjusting logging or testing parameters.

Installation

To set up and run the project, follow these steps:

1. Download and Install Python

   • Ensure you have Python 3.8 or higher installed. You can download it from python.org.

2. Set Up a Virtual Environment (Optional but Recommended)

   python -m venv venv
   source venv/bin/activate  # On macOS/Linux
   venv\Scripts\activate  # On Windows

3. Install Dependencies

   pip install -r requirements.txt

4. Run the LLM

   python main.py --local --save results.csv --num_ctx 32768 --file_logs logs.log --random_testing_dataframe 20 --save_interval 10

Command Line Arguments

When running main.py, the following arguments are available:

Argument	Short	Type	Description
--custom_client	-cc	str	Uses a custom client with the specified IP address.
--local	-l	flag	Runs using a local Ollama instance.
--save	-s	str	Specifies the file path where the results should be stored.
--num_ctx	-ctx	int	Sets a custom number of tokens for Ollama.
--file_logs	-fl	str	Saves logs to the specified filename instead of stdout.
--random_testing_dataframe	-rtd	int	Runs on a randomized subset of the dataset of the specified size.
--save_interval	-si	int	Saves results to the CSV file at the specified interval.

Usage

Main Program

Run the main script to classify articles using zero-shot prompts:

python main.py --local --save results.csv --num_ctx 32768 --file_logs logs.log --save_interval 50

Evaluation

To run the full evaluation of the classifier, execute:

python src/evaluation/full_eval.py

BERT Baseline

For comparison purposes, a BERT-based baseline classifier is provided. Run it with:

python src/bert/bert.py

License

This project is licensed under the MIT License. See the LICENSE file for details.