Design of Spam Detection and Classification System Based on Artificial Intelligence
Long Zhang, Wanni Liu, Jialong Wang · 2025
This article proposes a spam detection system based on multimodal deep learning and attention mechanism, which integrates an improved BERT wwm model, transfer learning, and Gated Attention Network (GAFN) to dynamically integrate multisource features such as text and links to optimize classification performance. The system was evaluated on three datasets-Enron, SpamAssassin, and a custom Chinese corpus-demonstrating significant superiority over baseline models such as BERT+CNN in F1 score ($94.6 \%$), AUC-ROC (0.962), false positive rate ($3.1 \%$), and latency (22.3 ms). After incorporating incremental learning, the detection rate of the new phishing email reached $89.7 \%$, with a false positive rate of only $3.8 \%$. Experimental results demonstrate that the system effectively balances high precision and real-time processing capabilities, providing an effective solution for dynamically combating spam emails and promoting the application of multimodal deep learning in the field of network security.