A 1D Convolutional Neural Network for Spam Classification

Emails and short messages are favoured for their communication efficiency. However, a substantial volume of spam is communicated even with recipient consent. To tackle such problems, models like TextCNN are employed for spam detection due to their simple network structures, swift training speed, and commendable performance. While TextCNN excels in spam classification, it has several drawbacks. For example, it utilized multi-channels with limited improvement and reduced explainability. It also lacks the capability to extract long-range features. Therefore, we introduce a novel 1D Convolutional Neural Network, overcoming TextCNN limitations by removing channels and employing an extended kernel. We assess both the proposed and the existing TextCNN models on a large-scale mixed dataset. The empirical results indicate a superior performance of the proposed model over TextCNN with more efficient computational cost, marking a significant advancement in spam classification.