DOI
https://doi.org/10.25772/67FQ-GN52
Defense Date
2024
Document Type
Thesis
Degree Name
Master of Science
Department
Computer Science
First Advisor
Bridget McInnes
Abstract
This thesis investigates the application of Few-Shot Learning (FSL) using Model-Agnostic Meta-Learning (MAML) to enhance Named Entity Recognition (NER) within the domain of Natural Language Processing (NLP), specifically focusing on chemical datasets. The primary challenge addressed is the impracticality of relying on extensive annotated datasets, especially in specialized fields like chemistry. The research primarily explores the concept of Few-Shot Learning, aiming to train models on minimal data while maintaining performance across diverse tasks. It delves into the N-way K-shot methodology, where "N" represents the number of classes and "K" signifies the number of examples per class. This approach is further investigated through the MAML method, a meta-learning strategy enabling models to quickly adapt to new tasks using only a few training examples. Key contributions of the thesis include the development of a comprehensive methodological framework employing MAML for NER tasks within the chemical context, demonstrated through experiments conducted on the ChEMU dataset. The challenges associated with applying FSL in NER are systematically presented, and an innovative solution is proposed through the adoption of the MAML method. The findings suggest that while FSL may not consistently outperform traditional models with large datasets, it offers a compelling alternative in scenarios where data is limited. This has significant implications for future research in NLP applications, particularly in specialized domains like chemistry.
Rights
© The Author
Is Part Of
VCU University Archives
Is Part Of
VCU Theses and Dissertations
Date of Submission
5-9-2024