Leveraging Bret And Data Augmentation For Robust Classification Of IMRAD Section In Research Papers

Abstract

This thesis investigates the automatic classification of IMRAD (Introduction,Methods,Results,and Discussion)sections in scientific papers using machine learning.A novel dataset incorporating à" Related Work" label was developed and used to train à BERT model,à state-of-the-art deep learning model fornatural language processing.The research employed data augmentation and robust data preprocessing techniques,including outlier detection and cleaning,to enhance the model's performance.The results demonstrate the effectiveness of transfer learning and data augmentation in achieving high accuracy and generalization for IMRAD classification. This work contributes to the field of sientific text processing by providing a robust framework for automated IMRAD classification and à valuable new dataset for future research.