Enhancing Android Malware Detection: The Influence of ChatGPT on Decision-centric Task

We explore the transformative impact of non-decision models, specifically ChatGPT, on the traditional decision-centric task of Android malware detection. Through a series of carefully designed experiments using publicly available datasets, this study reveals a paradigm shift. It reveals a serious lack of interpretability in decision-driven solutions, raising concerns about their reliability. In contrast, ChatGPT, as a non-decision-making model, is good at providing comprehensive analysis reports and significantly enhances interpretability. We give developers more insights through a non-decision-making perspective.

ChatGPT

You can find ChatGPT’s analysis report on [APK_Analysis].

Project Structure

• APK List: It contains the SHA256 of malicious and benign samples.

Dataset

All samples we used in our experiments you can find at [kronodroid].

Survey Results

We collect responses from 101 participants and process their data by removing personal or sensitive information.

Data preparation includes:

• Converting speech to text
• Translating Chinese responses to English
• Removing redundant modal particles

These processes ensure that the data is clean and structured, allowing for accurate and efficient analysis.