Emoji-Driven Sentiment Analysis for Social Bot Detection with Relational Graph Convolutional Networks

Area of Science:

• Computer Science
• Artificial Intelligence
• Cybersecurity

Background:

• Malicious social bots threaten cybersecurity and online information integrity.
• Current bot detection methods often neglect the semantic and emotional content of emojis in user-generated text.
• Emojis offer valuable cues for distinguishing between authentic and automated social media activity.

Purpose of the Study:

• To propose ESA-BotRGCN, an emoji-driven, multi-modal framework for enhanced social bot detection.
• To integrate semantic enhancement, sentiment analysis, and multi-dimensional feature modeling for improved accuracy.
• To address the limitations of existing methods by incorporating emoji-based emotional and semantic cues.

Main Methods:

• Established emoji-text mapping and used GPT-4 for textual coherence, generating tweet embeddings via RoBERTa.
• Extracted seven sentiment-based features to quantify emotional expression differences between bots and humans.
• Employed an attention gating mechanism to fuse sentiment features with user descriptions, tweet content, and network topology using a Relational Graph Convolutional Network (RGCN).

Main Results:

• Achieved a superior accuracy of 87.46% on the TwiBot-20 benchmark dataset.
• Significantly outperformed existing baseline bot detection models.
• Validated the effectiveness of emoji-driven semantic and sentiment enhancement strategies in improving detection performance.

Conclusions:

• The proposed ESA-BotRGCN framework demonstrates superior performance in social bot detection.
• Integrating emoji semantics and sentiment analysis is crucial for robust bot detection systems.
• The multi-modal approach effectively models heterogeneous social network topology for enhanced cybersecurity.