Abstract:
"Email overload is a challenge in today's digital world. This dissertation explores the
development of an application that leverages machine learning to prioritize emails based on
their content. The application utilizes BERT-Uncased, a transfer learning technique, to analyse
email text and assign priority labels. By integrating with the Gmail API, the application can
automatically download emails, categorize them based on importance, and update labels within
Gmail.
The project prioritizes novelty by addressing the gap in research on content-based email
prioritization. It tackles the challenge of limited data by employing BERT-Uncased, which is
efficient for working with smaller datasets. The chosen research methodology involves transfer
learning and API integration, balancing effectiveness with real-world feasibility.
The proposed solution offers a viable approach for email prioritization, requiring less data
compared to traditional methods. However, limitations exist: Using smaller datasets might
introduce bias, and the application could be generalized by incorporating more diverse datasets.
This thesis contributes to the field by demonstrating the potential of machine learning for email
management. It highlights the effectiveness of transfer learning techniques like BERT-Uncased
for practical applications with limited data. The evaluation considers perspectives from domain
experts, technical specialists, and target users, providing a well-rounded assessment of the
project's strengths, weaknesses, and future directions."
