Seed Community Identification Framework for Community Detection over Social Media

Area of Science:

• Data Mining and Social Network Analysis
• Computational Social Science

Background:

• Social media platforms facilitate community formation based on shared interests.
• Existing data mining clustering techniques struggle with implicit user information and network structures for robust community detection.
• Integrating passive and silent users' data presents a significant challenge in social media community analysis.

Purpose of the Study:

• To propose a novel Social Theory based Influence Maximization (STIM) framework for enhanced community detection on social media.
• To effectively integrate implicit user features, including those from passive users, into community structuring.
• To improve the robustness and modularity of identified social media communities.

Main Methods:

• Developed the STIM framework combining user-generated content and profile information.
• Employed influence maximization techniques to identify and incorporate passive social media users.
• Clustered nodes based on graphical parameters like degree, similarity, reachability, modularity, and density, guided by social theory.
• Utilized Convolutional Neural Networks (CNNs) for analysis within the STIM framework.

Main Results:

• The STIM framework demonstrated improved community detection by identifying denser and more modular communities.
• Achieved approximately 93% modularity and 94% Normalized Mutual Information (NMI).
• Showcased a 2.23% improvement in modularity and a 5.69% improvement in NMI compared to benchmark approaches.

Conclusions:

• The STIM framework offers a robust, social theory-driven approach to community detection in social media.
• The integration of influence maximization effectively addresses the challenge of incorporating passive user data.
• STIM provides a significant advancement in creating structurally sound and functionally relevant social media communities.