Research map / Reinforcement learning / Visual and representation RL
Towards Effective Planning Strategies for Dynamic Opinion Networks
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes a ranking-algorithm-labeled GCN classifier and an RL-based deep value network planner with five reward variants to choose intervention nodes that limit misinformation spread in dynamic opinion networks.
Problem
Intervention planning to disseminate accurate information and contain misinformation in dynamic opinion networks; exhaustive label generation for node-blocking search becomes computationally intractable as networks grow, and continuous opinion/trust models are under-explored.
Approach
A brute-force ranking algorithm generates labels for training GCN classifiers that pick nodes to block, and a deep value network trained with RL (experience replay) avoids costly label generation for larger networks; five reward functions and three opinion/trust cases are compared.
Key results
- On Dataset v2 (50 nodes, degree 4), the GCN trained on 10 nodes with RL+R1 reached an infection rate of 0.0842 at action budget 2 in Case-2, lower than ResNet(50) and the SL baseline (0.1032).
- Reward R3 (time-to-resolution) gave the highest infection rates in many settings, while single-factor rewards R1 and R4 performed best; R5 improved over R3.
- On real-world networks with budget 5, RL+R2 reduced infection on Zachary's Karate Club to 0.0662 from 0.0926 for R0.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use RL with GCN(10) instead of SL+GCN(25) | intervention planner · Dataset v2, 50 nodes, degree 4, Case-1, action budget 1 | infection rate: 0.304 -> 0.1593 | Helped |
| Use GCN(10) instead of ResNet(50) | RL planner · Dataset v2, 50 nodes, degree 4, Case-1, action budget 1 | infection rate: 0.199 -> 0.1593 | Helped |
| Add candidate-node count to the blocking-time reward (R3), forming R5 | RL planner · Dataset v2, 50 nodes, degree 4, Case-1, action budget 1, GCN(10) | infection rate: 0.2958 -> 0.1607 | Helped |
| Use the blocking-time reward R3 instead of candidate-node reward R1 | RL planner · Dataset v2, 50 nodes, degree 4, Case-1, action budget 1, GCN(10) | infection rate: 0.1608 -> 0.2958 | Hurt |
| Increase action budget from 1 to 3 | RL+R1, GCN(10) · Dataset v2, 50 nodes, degree 4, Case-1 | infection rate: 0.1608 -> 0.0597 | Helped |
| Use global infection-rate reward R4 instead of candidate-node reward R1 | RL planner · Dataset v2, 50 nodes, degree 4, Case-1, action budget 1, GCN(10) | infection rate: 0.1608 -> 0.1593 | Helped |
Surprising findings
- The sparse time-based reward R3 underperformed simpler candidate-node and infection-rate rewards, despite being cheapest to compute.
- A GCN trained on 10-node graphs generalized to 50-node test networks and outperformed a ResNet trained on 50-node graphs.
Limitations
- Centralized planners only; distributed planning not explored.
- Single-topic models without stubbornness or directed trust.
- Label generation via brute force is still used for SL, which is infeasible for large networks.
Experimental setup
Datasets: synthetic Watts-Strogatz networks Zachary's Karate Club Facebook Email Cora
Benchmarks: Dataset v2 (synthetic, 50 nodes, connectivity degree 4) Zachary's Karate Club Facebook Email Cora
Metrics: infection rate training MSE loss
Baselines: SL+GCN(25) RL+R0 RL+R3 ResNet(50)
Builds on: DQN DeGroot model GCN ResNet
Closest papers by meaning
- Expressive and Flexible Simulation of Information Spread Strategies in Social Networks Using Planning (AAAI 2024 · similarity 0.87)
- Sample Complexity of Linear Regression Models for Opinion Formation in Networks (AAAI 2025 · similarity 0.85)
- Opinion Maximization in Social Networks by Modifying Internal Opinions (NeurIPS 2025 · similarity 0.84)
- Amortized Network Intervention to Steer the Excitatory Point Processes (ICLR 2024 · similarity 0.84)
- Network Intervention by Polling Strategic Agents (NeurIPS 2026 · similarity 0.83)
- Graph Neural Network Based Action Ranking for Planning (NeurIPS 2025 · similarity 0.82)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.