E-Mail Spam Detection Using BERT and LSTM

The quick advancement of today's digital world and our daily reliance on the internet have a big impact on our lives. Especially in these days when automation is very easy, fraud and fraudulent activities against users have become quite common. One of the most common examples of these activities is fraud attempts through spam e-mails. Spam e-mails aim to hijack people's information, spread malware and defraud people by playing with their perceptions. Various spam detection methods and algorithms have been developed to prevent these problems. In this study, two models are proposed to detect spam e-mails using natural language processing (NLP) and deep learning methods: LSTM (Long Short Term Memory) and BERT (Bidirectional Encoder Representations from Transformers). As a result of the studies on the Enron e-mail dataset, the BERT model showed 98.93% accuracy and 98.87% f-score performance. The LSTM model showed 97.47% accuracy and f-score performance. With these results, it is seen that the BERT model detects spam with higher accuracy despite the complexity in language structures.