Multiscale Information Diffusion Prediction With Minimal Substitution Neural Network.

Area of Science:

• Social Network Analysis
• Computational Social Science
• Information Science

Background:

• Information diffusion prediction is complex due to content substitution dynamics on social platforms.
• Existing models often neglect factors like inter-information competition, user appeal, and content anticipation.
• Predicting information spread requires understanding both macroscopic popularity and microscopic user-level cascades.

Purpose of the Study:

• To propose a novel neural network model for multiscale information diffusion prediction.
• To simultaneously address macroscopic popularity and microscopic diffusion prediction tasks.
• To incorporate factors like content lifecycle, user preferences, and anticipation into diffusion modeling.

Main Methods:

• Developed a multiscale information diffusion prediction with minimal substitution (MIDPMS) neural network.
• Modeled information diffusion as a substitution system, integrating content lifecycle, user preferences, and anticipation.
• Utilized a minimal-substitution-theory-based neural network for joint macro/micro prediction training.

Main Results:

• The MIDPMS model demonstrated strong performance on both macroscopic and microscopic prediction tasks.
• Experiments on Weibo and Twitter datasets validated the model's effectiveness.
• The proposed approach successfully captured the complexities of information diffusion dynamics.

Conclusions:

• The MIDPMS model offers a robust solution for multiscale information diffusion prediction.
• Considering substitution dynamics and user anticipation enhances prediction accuracy.
• The model provides valuable insights for understanding and managing information spread on social media.