Full-Scale Information Diffusion Prediction With Reinforced Recurrent Networks.

Area of Science:

• Computational Social Science
• Machine Learning
• Network Science

Background:

• Information diffusion prediction models user behavior in information spread.
• Recurrent Neural Networks (RNNs) excel at sequential data but typically focus on either microscopic (next user/time) or macroscopic (total users) prediction.
• A unified model for both scales has been lacking.

Purpose of the Study:

• To propose a novel, unified model for full-scale information diffusion prediction.
• To integrate microscopic and macroscopic diffusion prediction capabilities.
• To leverage social graph structures for enhanced prediction accuracy.

Main Methods:

• Developed a full-scale diffusion prediction model using Reinforcement Learning (RL).
• Incorporated macroscopic diffusion size into an RNN-based microscopic model by addressing non-differentiable issues.
• Employed a structural context extraction strategy utilizing social graph information.

Main Results:

• The proposed RL-based model demonstrated superior performance compared to state-of-the-art baselines.
• Achieved significant improvements in both microscopic and macroscopic information diffusion predictions.
• Validation performed on three real-world datasets confirmed model effectiveness.

Conclusions:

• The novel RL model provides a unified approach for full-scale information diffusion prediction.
• Effectively integrates diverse prediction scales and social network structures.
• Represents a significant advancement over existing specialized models.