DGA Domain Name Detection Model Based on Gated Convolution and LSTM

• Kui Jiang,
• Siwei Wu,
• Ruibin Huang,
• Zhaorui Deng,

Vol. 19, No. 3, March 31, 2025
10.3837/tiis.2025.03.015, Download Paper (Free):

• Attention Mechanism
• Domain name generation algorithm
• Gated Convolution
• residual structure

Abstract

Malware employs domain name generation algorithms (DGA) to generate domain names, enabling communication with command and control servers for carrying out network attacks. This paper presents a novel domain name detection model for DGA, combining gated convolutional and LSTM(Long Short-Term Memory) architectures. The model consists of convolutional and LSTM layers, extracting local features and contextual relationships of domain names, respectively. Within the convolutional operation, a gating mechanism is introduced to control information flow and enhance transmission efficiency. By utilizing multiple convolutional gating units, deep-level features of domain names are extracted, while residual structures mitigate the issue of gradient dispersion. The LSTM layer incorporates an attention mechanism to capture crucial information from domain name characters, improving detection accuracy. Experimental results demonstrate the superiority of the proposed DGA domain name detection model compared to others.

Statistics

Cite this article