DNA-influenced automated behavior detection on twitter through relative entropy

Area of Science:

• Computer Science
• Network Science
• Computational Social Science

Background:

• Twitter's widespread use facilitates bot activity, impacting information integrity.
• Existing bot detection methods often overlook user profile heterogeneity and require extensive supervised learning data.
• There is a critical need for robust bot detection to combat misinformation and maintain online platform credibility.

Purpose of the Study:

• To propose a novel entropy-based framework for detecting correlated Twitter bots.
• To leverage user behavior patterns, modeled as DNA sequences, for bot detection.
• To overcome limitations of existing methods, such as reliance on topological structure and supervised learning.

Main Methods:

• Collected real-time Twitter user data.
• Modeled user online behaviors as DNA sequences.
• Computed relative entropy based on DNA sequence probability distributions to identify bots.

Main Results:

• The proposed entropy-based framework achieved high performance metrics: precision of 0.9471, recall of 0.9682, F1 score of 0.9511, and accuracy of 0.9457.
• Experimental results demonstrate superior performance compared to state-of-the-art bot detection techniques.
• The method effectively detects correlated bots using only user behavior data.

Conclusions:

• The novel entropy-based framework offers an effective and accurate approach for Twitter bot detection.
• By focusing on user behavior and employing an entropy measure, the method overcomes limitations of traditional techniques.
• This approach contributes to combating misinformation and enhancing the credibility of online social platforms.