Common Sense Reasoning Of Incomplete Instructions Based On Semantic Role Labeling And Reinforcement Learning

With the development of society,people have an increasing demand for service robots,especially home service robots.The problem of parsing incomplete natural language instructions into machine instructions through natural language processing has therefore become one of the research topics.The key to analysis is how to make the model recognize incomplete natural language instructions and infer the potential information of the instructions based on common sense.This thesis uses the current research status at home and abroad,with the indoor home environment as the background,through the combination of reinforcement learning,the analysis of incomplete natural language instructions is studied.First,select the corresponding instructions from the corpus as the training set and the test set,and label the training set instructions.Secondly,it optimizes some problems existing in the semantic role annotation model in the field of natural language processing.First,it normalizes by adding Layer Normalization between the layers of the semantic role annotation model based on the deep self-attention mechanism.Processing,and then introduce Highway Networks optimized two-way long and short-term memory neural network to optimize the recursive sub-layer in the model and other methods to accelerate the model's convergence speed,improve the model's robustness and model expression ability.Then,the instruction is parsed into a verb frame through the semantic role annotation of the deep self-attention mechanism,and the agent is trained to recognize the complete instruction through the reinforcement learning method.input the training set and use the Q-Learning algorithm to update the Q-table of the system agent,so that the agent can gradually "understand" the incomplete instruction.Finally,after completing the training work,start to use the test set to test,observe the test results,and adjust the parameters to improve the prediction accuracy.The feasibility and effectiveness of the method in this paper are verified through experiments,the experiments of this model under different conditions and the experiments of other papers in the same direction are compared,and the experimental results are analyzed.