Dynamic Malware Analysis using GPT-4 With 100% Recall Rate

A new Dynamic Malware Analysis model using GPT-4 and BERT overcomes quality issues in API call sequences in dynamic malware analysis. It performs better than the TextCNN method and generates explanatory texts for each API call. The texts are used to create representations, with features extracted by a new Convolutional Neural Network. The model can then connect with various malware code categories for further analysis. It improved on natural text representation and performed well on five benchmark datasets.