DQN-ESN

DRL-based approach integrating network embedding to address the competitive influence maximization on evolving social networks

Disclaimer

This work is based on existing GitHub Code (https://github.com/devsisters/DQN-tensorflow).

Competitive Influence Maximization on Dynmaic Social Networks Using DQN

Python Tensorflow implementation of Network Embedding Meets Deep Reinforcement Learning to Tackle Competitive Influence Maximization on Evolving Social Networks.

This implementation contains:

1. Deep Q-network
2. Experience replay memory

Requirements

• Python 2.7 or Python 3.3+
• gym
• tqdm
• SciPy or OpenCV2
• TensorFlow 0.12.0

Usage

First, install prerequisites with:

$ pip install tqdm gym[all]

Second, generate embeddings of a social graph (evoling social networks) using generate_embeddings.py inside data folder

Train a DQN Model on Evolving Social Networks

To train a model for an evolving social networks such as bitcoinalpha and bitcoinotc against opponent's degree strategy:

$ python main.py --env_name=bitcoinalpha --is_train=True --opponent degre
$ python main.py --env_name=bitcoinotc--is_train=True --opponent degre

Test a DQN Model on Evolving Social Networks

To test a model for an evolving social networks such as bitcoinalpha against opponent's weight strategy:

$ python main.py --is_train=False --env_name=bitcoinalph --opponent weight --testing_episode 2000