Vietnamese SMS Spam Detection with Deep Learning and Pre-trained Language Model

Abstract

Despite of the strong development of OTT message applications and social networks, Short Service Message (SMS) keeps its role in the marketing industry. As a top level of effective and cost-saving advertising tool, SMS has also given rise to SMS spam. To contribute for the fight against SMS spam, we suggested a model which is the combination of deep learning neural network model and pre-trained language technique – PhoBERT, a variant of BERT. Making full usage of the pre-training Vietnamese data, the proposed model achieved good accuracy at 99.53% in detecting Vietnamese spam messages.