Dataset Card for Polarization GameNet
Dataset Summary
Polarization GameNet is a benchmark dataset for modeling two-player zero-sum polarization games in social networks. Each data sample includes a synthetic social graph, node-level opinion states, and the ground truth Nash equilibrium strategies for both players, computed using a game-theoretic solver. The Nash equilibrium arises from a strategic interaction between two agents—one aiming to maximize network polarization and the other seeking to minimize it. This setting models adversarial and defensive interventions in public opinion, with broad relevance to research in algorithmic game theory, social influence modeling, and robust AI planning.
- ** Polarization GameNet enables tasks such as equilibrium strategy prediction, intervention policy learning, and robustness analysis under graph perturbations.
- ** We provide baseline solvers, evaluation metrics, and PyTorch-compatible loaders to support experimentation. The dataset and code are publicly available, and all instances are fully reproducible.
- ** This dataset is designed for researchers in graph learning, social dynamics, and algorithmic game theory.
Supported Tasks and Leaderboards
Supported Tasks
- Graph-based Nash equilibrium prediction
- Policy learning under networked strategic interaction
- Robustness evaluation of intervention strategies
Benchmark Leaderboards
No official leaderboard is hosted yet, but baseline performance metrics and evaluation scripts are included in the repository.
Dataset Structure
Each data sample includes:
adjacency: adjacency matrix of the network (shape: n x n)opinions: initial opinions on each node (shape: n x 1)player_1_strategy: Nash-optimal strategy vector for the polarizing agent (shape: n,)player_2_strategy: Nash-optimal strategy for the depolarizing agent (shape: n,)nash_value: value of the game at equilibrium (float)polarization_score: polarization level before the game (float)
Data Generation
- Networks are synthetically generated using Barabási–Albert, Watts–Strogatz, and Erdős–Rényi models.
- Initial opinions are drawn from a uniform distribution over [0, 1].
- The Nash equilibrium is computed using a custom solver designed for polarization games.
Intended Uses
This dataset is ideal for:
- Learning Nash equilibria in network games
- Modeling influence operations on social networks
- Studying adversarial and defensive interventions in opinion dynamics
- Benchmarking robustness and generalization of graph learning models
Dataset Creation
- Author: [xxx]
- Institution: [xxx]
- Year: 2026
This dataset includes three real-world networks, as well as purely synthetic networks that do not contain real user or social media data.
Licensing Information
This dataset is licensed under the MIT License. You are free to use, modify, and distribute it with attribution.
Citation
Coming soon (NeurIPS 2025 submission).
Contributions
We welcome community contributions of new baseline models, evaluation scripts, or alternative equilibrium solvers. Please open a pull request or file an issue in the GitHub repo.
- Downloads last month
- 7