Abstract: To enhance students' practical and research abilities, this paper designs an experimental teaching case for cross-language malicious code detection based on Semantic Enhanced Abstract Syntax Tree (SEAST). The experiment aims to construct the SEAST to retain the semantic structure of the code while eliminating non-semantic noise, achieving a unified representation of cross-language. At the same time, it incorporates a multi-branch Softmax classification mechanism to address issues of data distribution differences and overfitting in small sample languages. The experimental design covers the construction of a large-scale dataset, ablation studies, comparison experiments with state-of-the-art models, and verification experiments for detecting unknown languages, all addressing practical challenges. The comprehensive experimental process allows students to gain a deep understanding of the core technologies in malicious code detection, master experimental methods using deep learning techniques to solve network security problems and helps foster students' innovative thinking and practical abilities.