Deep Learning-Based Model for Detecting Fraudulent SMS Messages

With the rapid development of the Internet and the popularity of personal communication devices, smart phones have become a must-have for the public. While SMS is widely used for communication, a variety of messages are also sent to users through different channels. While bringing convenience to users, fraud methods for SMS as the carrier also emerge in an endless stream. In order to accurately identify scam short messages, this paper proposes a BERT-BiLSTM-EAA model. In the experiments, the proposed model not only outperforms the traditional machine learning models (such as support vector machine and logistic regression), but also outperforms other similar baseline models that are also based on bert architecture or introduce an attention mechanism on the fraud SMS dataset. The precision of the model reaches 93.98%, the precision reaches 87.36%, the recall reaches 84.32%, and the f1 score reaches 85.18%. This shows that the model has made important progress in the field of identifying scam SMS messages.