A Sentence-Level Joint Relation Classification Model Based on Reinforcement Learning

Algorithm 1

Joint training of the RL model and joint network model.

	Input: Number of Episode N. Training data X, Initialize the RL model parameters and joint network model Parameters
	Output: RL model parameters ψ and joint network model Parameters θ
(1)	for episode n = 1 to N do
(2)	foreachdo
(3)	Calculate the predicted score for each state
(4)	According to the predicted score, the action taken on the state is obtained
(5)	Calculate temporary and average Awards
(6)	Update the parameters of RL model
(7)	Calculate total award
(8)	end foreach
(9)	Train and update the parameters θ of joint network model
(10)	Update the parameters of RL model
(11)	Find the best parameters for RL model according to the reward
(12)	Update the weights of the RL networks
(13)	end for