Topic Modelling Using BERTopic for Robust Spam Detection

Abstract

Spam emails continue to be a challenging issue in terms of cybersecurity. This study uses state-of-the-art techniques such as BERTopic and various text-mining strategies to effectively address this issue. The study also compared the performance of four different embedding models in topic modeling. In particular, the experimental results underline the outstanding performance of the 'Roberta-base' model and highlight its effectiveness in detecting spam emails. The importance of this study lies in its role in demonstrating the effectiveness of BERTopic and related methodologies in combating spam emails, providing valuable information to researchers and practitioners on email security, natural language processing, and machine learning. The study's comprehensive examination of spam datasets, combined with rigorous comparative analyses of placement patterns, advances the existing knowledge base in this area. © 2024 IEEE.