Research Article
IIDQN: An Incentive Improved DQN Algorithm in EBSN Recommender System
Algorithm 1
Incentive improvement algorithm based on Q networks in EBSN.
| IIDQN algorithm | | Inputs: state space , action space , discount rate , learning rate , parameter update interval . | (1) | Randomly initialize the parameters of the Q-networks and randomly initialize the parameters of the target Q-networks. | (2) | Initialization state . | (3) | Select action . | (4) | Perform action, get rewardand next actionthrough environment. | (5) | Put , , , into the experience pool and sample | (6) | Sample , , , in the experience pool for the next action. Let y be the target value | | | (7) | Update the weights by back propagation mechanism and retrain the Q networks using gradient descent algorithm | (8) | Detect sparse nodes and apply additional reward/penalty updates to nodes | | | (9) | Update the target networks every steps and copy the current networks parameters to the target networks. |
|