Combating QR-Code-Based Compromised Accounts in Mobile Social Networks.

Area of Science:

• Cyber Physical Social Sensing
• Mobile Social Networks (MSNs)
• Machine Learning

Background:

• Mobile social networks (MSNs) are vulnerable to attacks where malicious URLs spread via QR codes compromise accounts to propagate spam.
• Existing detection methods for compromised accounts in MSNs are platform-restrictive and overly simplistic, failing to address the unique interactions within these networks.

Purpose of the Study:

• To develop an effective method for tracing attacks to their source within MSNs to stop the spread of compromised accounts.
• To leverage sensor-related information, specifically GPS location data, to identify compromised accounts in MSNs.

Main Methods:

• Analysis of information sending modes between compromised and normal accounts.
• Utilizing Global Positioning System (GPS)-based location information regularity.
• Construction of an entropy-based model using machine learning strategies, incorporating entropy and conditional entropy concepts.

Main Results:

• The entropy-based model achieved an accuracy rate of 87.6% with a low false positive rate of 3.7%.
• Comparative experiments confirmed that sensor-based location information is a viable feature for detecting compromised accounts in MSNs.
• The study analyzed approximately 500,000 Sina Weibo accounts and 100 million messages.

Conclusions:

• The proposed entropy-based model effectively detects compromised accounts in MSNs by analyzing sensor data, particularly GPS location patterns.
• Leveraging sensor data, especially location information, offers a more robust and platform-independent approach to identifying malicious activities in MSNs.
• This research provides a significant advancement in securing mobile social networks against covert attacks originating from compromised user accounts.