Research On The Reasoning Method Of Agricultural Knowledge Graph Based On Proximal Policy Optimization

As one of the important means of contemporary massive network data storage,knowledge graph technology has been widely used in many fields such as agriculture,medical treatment and machinery manufacturing.However,the current knowledge graph still has shortcomings such as low level of automated construction and insufficient completeness.Therefore,reasoning and complementing the relationship in the knowledge graph is very important in the current research of knowledge graph technology.In the current research on knowledge graph reasoning problems,new methods are emerging one after another,and there is no recognized and mature knowledge reasoning method so far.Among them,the knowledge reasoning method based on the strategy gradient reinforcement learning algorithm has the best reasoning effect on the whole,but there is still a problem that the step length is difficult to determine in the training process,resulting in low training efficiency and insufficient training results.Over time,the new reinforcement learning algorithm has been able to solve the step size problem in the old algorithm.Therefore,it is of great theoretical significance and application value to integrate this advanced reinforcement learning algorithm into the knowledge reasoning process to obtain a better reasoning effect.Based on the existing reinforcement learning knowledge reasoning technology at home and abroad,this thesis uses the PPO reinforcement learning algorithm as the training method of the agent in the model for the first time,and proposes a "domain reward function" for graph reasoning in the agricultural domain.In the research process,the theory and experiment are combined,and at the same time,it is compared with a variety of other knowledge reasoning methods.The main research work and results completed in this thesis are summarized as follows:1)A knowledge graph reasoning method based on PPO reinforcement learning algorithm is proposed.This method mainly solves the problem that the step length is difficult to determine in the knowledge reasoning method based on the strategy gradient reinforcement learning algorithm.Firstly,this method is theoretically analyzed and algorithmic modeling,and secondly,the knowledge graph data set is processed,and the performance of four different reasoning methods under two reasoning tasks is analyzed to demonstrate the effectiveness of this method.2)Proposed the PPO-AGR algorithm,which designed a "domain reward function" for the agricultural data set based on the knowledge reasoning method based on the PPO algorithm above.First,analyze and process the characteristics of the agricultural data set,and define the concepts of root entity and root entity relationship.By analyzing the performance of five different reasoning methods(including the methods proposed above)under three reasoning tasks,the characteristics and differences of various methods are summarized.The results show that the method based on PPO effectively improves the accuracy and efficiency of reasoning on public data sets,and the method based on PPO-AGR further improves the reasoning effect in the reasoning of root entity relationships on agricultural data sets.