Evaluating the Performance of State-of-the-Art ESG Domain-Specific Pre-Trained Large Language Models in Text Classification Against Existing Models and Traditional Machine Learning Techniques - Master of Science Dissertation

Aug 6, 2026

Upload by Maeve

This dissertation evaluates state-of-the-art ESG domain-specific pre-trained large language models for text classification, positioning their effectiveness against existing modeling approaches and traditional machine learning techniques. The work examines ESG concepts and large language model foundations, then develops fine-tuning experiments, classifier training, and a comparative evaluation framework using performance metrics. Implementation, data collection and preprocessing, refinement cycles, and ethical considerations support the research design, hypotheses, and research objectives.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages58

Reading Time2 h 26 min

AccessFree to download

Executive Summary

Introduction

Motivation and Background

Research Aim

Research Focus

Research Questions/Hypotheses

Research Objectives

Structure of the dissertation

Literature Review

Environment, Social, and Governance (ESG)

Large Language Model (LLM)

BERT (Bidirectional Encoder Representations from Transformers)

LLaMA2 (Large Language Model Meta AI 2)

Classical Machine Learning (ML), Support Vector Machine (SVM) and XGBoost

LLM Fine-Tuning

Research Gap

Methodologies and Method

Research Philosophy

Research Method/Methodology

Research Design

Implementation

Ethical Considerations

Department of Computer Science EVALUATING THE PERFORMANCE OF STATE-OF-THEART ESG DOMAIN-SPECIFIC PRE-TRAINED LARGE LANGUAGE MODELS IN TEXT CLASSIFICATION AGAINST EXISTING MODELS AND TRADITIONAL MACHINE LEARNING TECHNIQUES Module Code: COM00151M Module Name: Independent Research Project by Tin Yuet CHUNG June 2024 supervised by Dr. Majid Latifi word count: 9395 A dissertation submitted in partial fulfilment of the requirement for the degree of Master of Science in Computer Science with Artificial Intelligence Acknowledgement First of all, I would like to thank Dr. Majid Latifi for his supervision and help throughout the duration of the project. In the last few months, Majid has provided me the direction of research and has given insightful comments. I also want to thank Tobias Schimanski and his research group who contributed to the ESG annotated text data. Secondly, I would like to thank my friends in Hong Kong and Canada. Last but not least, I would like to thank my family for supporting me in Canada. Table of Contents Acknowledgement ................................................................................................................. 0 List of Figures ........................................................................................................................ 3 Lists of Tables ........................................................................................................................ 3 Table of Abbreviations ............................................................................................................ 4 Executive Summary ............................................................................................................... 5 1- Introduction ....................................................................................................................... 7 1.1 Motivation and Background ........................................................................................... 7 1.2 Research Aim ............................................................................................................... 7 1.3 Research Focus ............................................................................................................ 8 1.4 Research Questions/Hypotheses .................................................................................. 8 1.5 Research Objectives ..................................................................................................... 8 1.6 Structure of the dissertation .......................................................................................... 8 2-Literature Review ............................................................................................................. 10 2.1 Environment, Social, and Governance (ESG) ................................................................ 10 2.2 Large Language Model (LLM) ....................................................................................... 10 2.3 BERT (Bidirectional Encoder Representations from Transformers) ................................ 11 2.4 LLaMA2 (Large Language Model Meta AI 2) ................................................................... 12 2.5 Classical Machine Learning (ML), Support Vector Machine (SVM) and XGBoost............. 13 2.6 LLM Fine-Tuning.......................................................................................................... 14 2.6.1 What is LLM Fine-Tuning? ..................................................................................... 14 2.6.2 Advantages and real-world applications of fine-tuned models ............................... 15 2.6.3 LLM Fine-Tuning Techniques ................................................................................. 15 2.6.4 Qlora.................................................................................................................... 16 2.7 Research Gap ................

FAQs

What is the main objective of the dissertation?

The dissertation evaluates how well ESG domain-specific pre-trained large language models perform for text classification compared with existing models and traditional machine learning techniques.

Which model families and approaches are reviewed and compared?

The literature review covers ESG, LLMs, BERT, LLaMA2, classical ML methods including SVM and XGBoost, and LLM fine-tuning, including QLoRA.

How is the research carried out and evaluated?

The methodology includes fine-tuning experiments, classifier training, data collection and preprocessing, and performance evaluation metrics, followed by refinement and re-evaluation through iterative cycles.

Aug 6, 2026

Upload by Maeve