Abstract
As computers are widely used in people's work and daily lives, malware has become an increasing threat to network security. Although researchers have introduced traditional machine learning and deep learning methods to conduct extensive research on functions in malware detection, these methods have largely ignored the analysis of function parameters and functional dependencies. To address these limitations, we propose a new malware detection method. Specifically, we first design a parameter encoder to convert various types of function parameters into feature vectors, and then discretize various parameter features through clustering methods to enhance the representation of API encoding. Additionally, we design a deep neural network to capture functional dependencies, enabling the generation of robust semantic representations of function sequences. Experiments on a large-scale malware detection dataset demonstrate that our method outperforms other techniques, achieving 98.62% accuracy and a 98.40% F1-score. Furthermore, the results of ablation experiments show the important role of function parameters and functional dependencies in malware detection.