Link Prediction Based on Stochastic Information Diffusion.

Area of Science:

• Network Science
• Machine Learning
• Computational Social Science

Background:

• Link prediction (LP) is crucial for understanding future interactions in diverse networks like e-commerce and social media.
• Existing LP methods often overlook network information flow and dynamic processes, limiting their predictive power.
• Information diffusion significantly influences link formation but remains an under-explored area in LP research.

Purpose of the Study:

• To develop a novel link prediction method that incorporates information diffusion dynamics.
• To propose a new evaluation metric that assesses both diffusion capacity and prediction accuracy.
• To demonstrate the effectiveness of the proposed method against existing techniques.

Main Methods:

• Analysis of link prediction effects using susceptible-infected-recovered and independent cascade diffusion models.
• Development of the Progressive-Diffusion (PD) method, a stochastic discrete-time rumor model based on node propagation dynamics.
• Introduction of a novel evaluation metric combining information diffusion and link prediction accuracy.

Main Results:

• The Progressive-Diffusion (PD) method demonstrates superior performance in link prediction compared to prior art.
• The PD model exhibits low-memory and low-processing requirements, suitable for parallel and distributed systems.
• Experimental results validate the effectiveness of the proposed method and evaluation metric.

Conclusions:

• Incorporating information diffusion dynamics significantly enhances link prediction accuracy.
• The Progressive-Diffusion (PD) method offers an efficient and effective approach for dynamic network analysis.
• The novel evaluation metric provides a comprehensive assessment of link prediction methods in diffusion contexts.