A BERT-GCN-Based Detection Method for FBS Telecom Fraud Chinese SMS Texts
Xuanming Zhang, Ruiyang Huang, Lixiao Jin, Fangjie Wan · 2023
In the past decade, Fake Base Stations (FBS) have been consistently employed by criminals to target mobile users through spam text messages. Despite the introduction of several techniques to mitigate this problem, spam messages remain a persistent and challenging issue in some countries, such as China, resulting in billions of dollars in annual economic losses. Therefore, this paper proposes an algorithm named BERT-GCN that combines large-scale pre-training models with Graph Convolutional Networks (GCN) for multi-class detection of fraudulent base station telecommunications scams in Chinese text messages. First, BERT (Bidirectional Encoder Representation from Transformers) is used to encode the corpus and generate word embeddings. Subsequently, the generated word embeddings are fed into GCN for training. Finally, a Softmax layer is employed to perform multi-class detection on fraudulent base station telecommunications scams in Chinese text messages. Experimental results demonstrate that the proposed model achieves favorable performance on the FBS-SMS-Dataset and exhibits outstanding performance in the domain of multi-class detection.