This repository contains code for our paper titel: 'Enhancing Relation Extraction Through Augmented Data: Large Language Models Unleashed' The paper is accepted at NLDB2024. Both the augmented data and the original dataset are available. All the code is self-explanatory, and we have provided it in notebook format for easy understanding. Most of the cells contain instructions.

All the packages used in our experiments are easy to install with the pip command. If you encounter any issues related to packages or code execution, please create an issue, and we will be happy to assist and resolve the problem.

The augmented data from our experiments is available below:

• The "FewRel" directory contains all the augmented data generated using both the large language model and the rule-based approach.
• The "NYT-FB" directory contains data organized according to the different schemes explained in the paper.

We provide complete code in the notebook for easy execution and step-by-step understanding. By default, we've set Llama-2-7B version, which is easy to run. Users can easily change it to another version in the model name cell.

Llama-Based Augmentation

Similar to Llama, the code for augmenting data using Falcon is in the following notebook.

Falcon-Based Augmentation

The following notebook contains code for generating rule-based augmentation following the same approach as we followed for the two language models.

Rule-Based Augmentation

The choosen model for evaluation and it's relvant code for training and evaluation is available in

Model

• Manzoor Ali (DICE, Paderborn University)
• Muhammad Sohail Nisar (DICE, Paderborn University)
• Muhammad Saleem (DICE, Paderborn University)
• Diego Moussallem (DICE, Paderborn University)
• Axel-Cyrille Ngonga Ngomo (DICE, Paderborn University)

The source code of this repo is published under the GNU General Public License v3.0