- **Tested OS: **Linux
- Python >= 3.8
- PyTorch == 1.13.0
- Tensorboard
- Install PyTorch 1.13.0 with the correct CUDA version.
- Set the following environment variable to avoid problems with multiprocess trajectory sampling:
export OMP_NUM_THREADS=1
You can train your own models using the provided config in metro/cfg
:
python -m news.train --cfg cfg_name --global_seed 0 --num_threads 1 --gpu_index 2 --agent rl-gnn3
You can replace cfg_name
to train other cfgs.
The results are saved in path result/platform/method/cfg/seed
- We compared DRLE to other baselines at different social media platforms, with the metrics being the Total Infectious Rate.
- Extensive experiments demonstrate that DRLE yields impressive effects on the mitigation of rumors, exhibiting an improvement of over 20% compared to baseline methods.
- The model trained on small social media platforms can be directly applied to larger networks with only a marginal decrease in metrics. Importantly, this performance remains superior to the optimal baselines.
- DRLE can also offer effective protection for specific populations within social media platforms.
Please see the license for further details.
The implemention is based on Transform2Act.