Document-Level Entity Relation Extraction

This GitHub repository contains the codebase and documentation for a comprehensive study on document-level entity relation extraction, conducted as part of a thesis. The research focuses on leveraging state-of-the-art generative models, with a primary emphasis on fine-tuning the T5 model and exploring the capabilities of GPT-3.5 Turbo and GPT-4 in zero-shot and one-shot settings.

Overview

Entity relation extraction plays a crucial role in natural language processing, and while traditional approaches concentrate on sentence-level analysis, this research delves into the emerging domain of document-level entity relation extraction. The complexity arises from dependencies and connections that go beyond individual sentences, presenting unique challenges in understanding relationships within a broader context.

Key Features

• Exploration of generative models for document-level entity relation extraction.
• Fine-tuning of the T5 model and its variants for capturing nuanced relationships.
• Evaluation of GPT-3.5 Turbo and GPT-4 in zero-shot and one-shot scenarios.
• Comparative analysis of T5 and GPT models for document-level extraction performance.

How to Use

1. Clone the repository:

   git clone https://github.com/dunja274/entity-relation-extraction.git
   cd entity-relation-extraction

2. Install dependencies:

   pip install -r requirements.txt

3. Run experiments and evaluate models using provided scripts.

Requesting Full Thesis and Results

This repository presents a condensed version of the thesis. For a detailed understanding of the research methodology, results, and conclusions, you can request the full thesis by sending an email to your@email.com.

Thesis Structure

• Chapter 2: Review of related work in entity relation extraction.
• Chapter 3: Insights into the utilized dataset.
• Chapter 4: Task modeling, experimental setup, and prompt engineering.
• Chapter 5: Discussion of models, including T5 variants and GPT models.
• Chapter 6: Evaluation metrics, including ROUGE-1, ROUGE-2, and Rouge-L scores.
• Chapter 7: Analysis of experiments and model performances.
• Chapter 8: Conclusions, limitations, and suggestions for future work.

Note: This project is a part of an academic thesis, and the full thesis can be requested for a more in-depth exploration of the research and its results.

Keywords: Natural Language Processing, Entity Relation Extraction, Document-Level Analysis, T5 Model, GPT Models, Generative Approaches, Information Extraction.

Thesis Outline

1. Introduction

   • Brief overview of the research topic.

2. Related Work 2.1. Early Approaches

   • Discussion of early approaches in entity relation extraction. 2.2. Deep Learning Approaches
   • Overview of deep learning approaches in entity relation extraction.

3. Dataset 3.1. Preprocessing

   • Details on dataset preprocessing. 3.2. Statistics
   • Presentation of dataset statistics.

4. Task Modeling 4.1. Relation Extraction as Extractive Summarization

   • Exploration of relation extraction as extractive summarization. 4.2. Experiments
     4.2.1. Named Entity Recognition and Relation Extraction
     • Details on experiments involving named entity recognition and relation extraction. 4.2.2. Tags and Relation Extraction
     • Overview of experiments involving tags and relation extraction. 4.2.3. Entities and Relation Extraction
     • Experiments related to entities and relation extraction. 4.3. Prompting
       4.3.1. Named Entity Recognition and Relation Extraction Prompt Design
     • Design of prompts for named entity recognition and relation extraction. 4.3.2. Relation Extraction Prompt Design
     • Design of prompts for relation extraction. 4.3.3. Entities and Relation Extraction Prompt Design
     • Design of prompts for entities and relation extraction.

5. Models 5.1. Text-to-Text Transfer Transformers
   5.1.1. Multilingual T5
   - Discussion on multilingual T5. 5.1.2. FLAN-T5
   - Overview of FLAN-T5. 5.2. Generative Pre-trained Transformers
   5.2.1. GPT-3.5 Turbo
   - Details on GPT-3.5 Turbo. 5.2.2. GPT-4
   - Overview of GPT-4. 5.3. Comparison of T5 and GPT models

   • Comparative analysis of T5 and GPT models.

6. Metrics 6.1. ROUGE-1 Score

   • Explanation of ROUGE-1 score. 6.2. ROUGE-2 Score
   • Explanation of ROUGE-2 score. 6.3. ROUGE-L Score
   • Explanation of ROUGE-L score. 6.4. Comparison of ROUGE Scores
   • Comparative analysis of ROUGE scores. 6.5. Comparison Between ROUGE and other Similarity Scores
   • Comparative analysis between ROUGE and other similarity scores.

7. Results and Discussion 7.1. T5 Results
   7.1.1. Analysis of Model Performance and Experimental Configurations
   - Discussion on T5 model performance and experimental configurations. 7.1.2. Analysis of Model Performance and Dataset Size
   - Discussion on T5 model performance concerning dataset size. 7.2. GPT Results

   • Discussion on GPT model results.

8. Conclusions

Bibliography