Deep reinforcement learning-based approach for rumor influence minimization in social networks

Area of Science:

• Social Network Analysis
• Information Diffusion
• Computational Social Science

Background:

• Malicious rumors on social networks can cause significant societal disruption, political conflict, and public opinion manipulation.
• Existing rumor influence minimization (RIM) strategies often rely on static approaches, failing to capture the dynamic nature of rumor propagation.
• Deep reinforcement learning offers a promising avenue for addressing the complexities of rumor evolution.

Purpose of the Study:

• Introduce the dynamic rumor influence minimization (DRIM) problem as a discrete-time optimization challenge.
• Propose a novel deep reinforcement learning-based dynamic rumor-blocking approach (RLDB).
• Enhance rumor propagation simulation accuracy through a dynamic rumor propagation model (DRPM).

Main Methods:

• Developed static rumor propagation model (SRPM) and dynamic rumor propagation model (DRPM) based on independent cascade patterns.
• Implemented RLDB, a deep reinforcement learning strategy that dynamically identifies users for blocking based on network dynamics and user states.
• Validated the RLDB approach using four diverse real-world social network datasets.

Main Results:

• The dynamic rumor propagation model (DRPM) dynamically adjusts probabilities, improving simulation accuracy.
• RLDB demonstrated superior performance in minimizing rumor influence compared to heuristic methods like out-degree, betweenness centrality, and PageRank.
• Experimental results confirmed the effectiveness of the proposed dynamic blocking strategy.

Conclusions:

• The RLDB approach provides an effective dynamic solution for rumor influence minimization in social networks.
• Dynamic modeling and deep reinforcement learning are crucial for addressing complex information diffusion challenges.
• The proposed method offers a significant advancement over traditional static rumor control strategies.