Detecting bots in social-networks using node and structural embeddings

Area of Science:

• Computer Science
• Social Network Analysis
• Machine Learning

Background:

• Social networks like Twitter enable anonymous interactions, facilitating bot accounts that mimic real users.
• Bot detection often relies on user profile metadata and natural language processing (NLP) features from tweets.
• Features derived from the underlying social network structure are less explored for bot detection.

Purpose of the Study:

• To investigate the effectiveness of network structure features for bot detection.
• To compare classical embedding techniques with structural embedding algorithms for identifying Twitter bots.
• To highlight the predictive power of local network structures around bot accounts.

Main Methods:

• Explored two classes of embedding algorithms: classical (proximity-based) and structural (local neighborhood structure).
• Applied these algorithms to extract features from Twitter's social network data.
• Evaluated the predictive performance of the extracted features for bot classification.

Main Results:

• Features derived from structural embeddings demonstrated higher predictive power in bot detection.
• Classical embedding techniques showed lower effectiveness compared to structural embeddings.
• The local social network structure around bot accounts is a valuable source for identification.

Conclusions:

• Structural embeddings offer a more potent approach to leveraging social network data for bot detection.
• Understanding local network patterns is crucial for improving the accuracy of automated bot identification systems.
• Further research into network-based features can enhance the robustness of bot detection models.