Pclf: Parallel cnn-lstm fusion model for sms spam filtering

Abstract

Short Message Service (SMS) is widely used for its accessibility, simplicity, and cost-effectiveness in communication, bank notifications, and identity confirmation. The increase in spam text messages presents significant challenges, including time waste, potential financial scams, and annoyance for users and carriers. This paper proposes a novel deep learning model based on parallel structure in the feature extraction step to address this challenge, unlike the traditional models that only enhance the classifier. This parallel model fuses local and temporal features to enhance feature representation by combining convolutional neural networks (CNN) and long short-term memory networks (LSTM). The performance of this model has been evaluated on the UCI SMS Collection V.1 dataset, which comprises both spam and ham messages. The model achieves an accuracy of 99.28% on this dataset. Also, the model demonstrates good precision, recall, and F1 score. This paper aims to provide the best protection from unwanted messages for mobile phone users.