iltc-narrative-classification

iltc-narrative-classification

Project Overview

This project is designed to process and classify narrative structures using deep learning techniques for Semeval2025 task 10-2. The main file to run is begin25.ipynb, which has been tested on Python 3.12.6 and requires Jupyter Notebook with a Python kernel. You can modify the variables within this file to customize the model according to your requirements. The project has been tested only on Windows 11.

Dataset Structure

The raw dataset file (in English), containing every sentence, must be placed in the following directory:
dataset/raw-documents from the root of the project. Additionally, you need to define the following three files within the dataset folder:

1. dataset/dev.txt

   • This file contains the dataset predictions to be sent to SemEval.
   • Each line should include a filename, and each line must be terminated with a newline character (\n).

2. dataset/test.txt

   • This file is used to calculate metrics like validation and threshold determination, which will later be applied to the dev.txt dataset.

3. dataset/train.txt

   • This file is used to define the training dataset.

There are present few created samples to help the setup of files in those folders.

File Format

Both test.txt and train.txt must follow the SemEval2025 file convention:

File.txt/tNarrative/tSubnarrative/n

Libraries needed

You need to install the following libraries:

tensorflow
keras
numpy
json
pickle
torch
matplotlib
transformers

Pre-trained networks

The used neural networks for final submissions are avaiable in this Google drive folder, and can be loaded from the main file, you can review them using keras summary functionalities.

Old Model

Through, uncommented and not reviewed,in 2024 folder its present the version of the model used for 2024 post competition.