AI-Enhanced Prediction of Collective Opinion Shifts and Cooperative Behavior Diffusion in Online Social Networks

Abstract

The interplay between individual opinion formation and cooperative behavioral choices in online social networks represents a critical yet underexplored dimension of computational social science. This study presents an AI-enhanced dual-layer network framework that integrates mathematical opinion-behavior co-evolution modeling with state-of-the-art machine learning techniques to predict collective opinion shifts and cooperative behavior diffusion. The opinion layer adopts a weighted DeGroot-like update rule, while the behavior layer synthesizes three social influence mechanisms—neighbor imitation, payoff-driven decision-making, and cognitive consistency pressure. Building upon this foundation, we develop a graph convolutional network (GCN) architecture with graph attention (GAT) aggregation to capture topological opinion propagation patterns, a proximal policy optimization (PPO)-based reinforcement learning agent for cooperative strategy optimization, and an ensemble of machine learning classifiers for opinion shift detection. SHAP (SHapley Additive exPlanations) analysis is applied to quantify the contribution of each network and behavioral feature to model predictions. Experiments across scale-free, random, and small-world network topologies reveal that scale-free structures accelerate opinion convergence and sustain higher cooperation ratios under both synchronous and asynchronous update mechanisms. The GCN-GAT model achieves a mean absolute error (MAE) of 0.031 for opinion prediction, outperforming baseline LSTM and random forest models. Empirical validation against longitudinal survey data on environmental attitudes and pro-environmental behaviors among Chinese social media users demonstrates strong concordance between model trajectories and observed trends. This work advances the understanding of opinion-behavior co-evolution by providing an interpretable, data-driven prediction framework applicable to public opinion monitoring, social intervention design, and behavioral policy assessment.

Keywords: Social networks; Opinion dynamics; Cooperative behavior; Graph neural networks; Reinforcement learning; SHAP explainability; Dual-layer network; Co-evolution