Towards Effective Planning Strategies for Dynamic Opinion Networks

TL;DR

The findings reveal that graph convolutional network-based planners facilitate scalable centralized plans that achieve lower infection rates (higher control) across various network configurations, including Watts-Strogatz topology, varying action budgets, varying initial infected nodes, and varying degrees of infected nodes.

Abstract

In this study, we investigate the under-explored intervention planning aimed at disseminating accurate information within dynamic opinion networks by leveraging learning strategies. Intervention planning involves identifying key nodes (search) and exerting control (e.g., disseminating accurate or official information through the nodes) to mitigate the influence of misinformation. However, as the network size increases, the problem becomes computationally intractable. To address this, we first introduce a ranking algorithm to identify key nodes for disseminating accurate information, which facilitates the training of neural network classifiers that provide generalized solutions for the search and planning problems. Second, we mitigate the complexity of label generation, which becomes challenging as the network grows, by developing a reinforcement learning-based centralized dynamic planning framework. We analyze these NN-based planners for opinion networks governed by two dynamic propagation models. Each model incorporates both binary and continuous opinion and trust representations. Our experimental results demonstrate that the ranking algorithm-based classifiers provide plans that enhance infection rate control, especially with increased action budgets for small networks. Further, we observe that the reward strategies focusing on key metrics, such as the number of susceptible nodes and infection rates, outperform those prioritizing faster blocking strategies. Additionally, our findings reveal that graph convolutional network-based planners facilitate scalable centralized plans that achieve lower infection rates (higher control) across various network configurations, including Watts-Strogatz topology, varying action budgets, varying initial infected nodes, and varying degrees of infected nodes.

Figures (47)

• Figure 1: Example of misinformation propagation and control choices at each timestep. Blue nodes: neutral (opinion value $0$), red nodes: misinformed (opinion value $-1$), green nodes: received accurate information (opinion value $1$).
• Figure 2: Sequence of actions chosen by the RL agent trained using reward function $R_3$.
• Figure 3: Comparative analysis of the GCN-Based SL Model Against Baseline Models Across Different Network Types and Budgets. Each subfigure represents one of the three cases $(1, 2,$ and $3)$, organized by rows, for three different types of networks: Tree, Erdős-Rényi, and Watts-Strogatz, organized by columns. Within each panel, the infection rate is plotted for four methodologies. SL based on GCN (blue), random node selection (orange), static selection of maximum degrees (green), and dynamic selection of maximum degrees (red) across three levels of budget (1, 2, and 3). These results underscore the variability in performance with changes in network structure and budget allocation, highlighting the superior effectiveness of the GCN model in simpler cases and under increased budget conditions, with diminishing returns in more complex environments.
• Figure 7: Case-1: Comparative MSE loss across different reward functions for a ResNet model trained on a 50-node dataset. Columns represent an increase in action budget during training, while rows indicate a rise in the number of initial infected nodes.
• Figure 8: Case-1: Comparative Mean Infection Rate across different reward functions for a ResNet model trained on a 50-node dataset tested on Dataset v2 of 50 nodes with degree of connectivity 3.