| Under the background of mobile Internet and big data,intelligent tutoring systems have gradually attracted people’s attention,and various intelligent tutoring systems have entered people’s lives.As the core of the intelligent tutoring system,machine answers can allow machines to understand and answer input questions,thereby effectively improving the quality of teaching and making up for the lack of teaching resources.The research on automatic machine answers is of great significance for promoting the development of machine intelligence and realizing intelligent education.The automatic solution of kinematics problems is a complex process of knowledge extraction,representation,reasoning,and calculation,integrating mathematics,linguistics,computer science,and other disciplines.Due to the lack of research on linguistic analysis and understanding of kinematics problems,it is necessary to study a high-performance algorithm to understand and solve kinematics problems.This paper divides the solving of kinematics problems into two stages: understanding and solving.For the understanding method of kinematics problems,the input text is converted into relational triples to represent the problem description.First,the pipeline architecture is used to solve the problem of understanding.Then a named entity recognition method is proposed based on the graph attention neural network and the Bi-LSTM network.Finally,a relationship extraction method between the graph attention neural network and the SelfAttention mechanism is proposed to extract entity-relationship pairs in kinematics problems,which lays the foundation for solving kinematics problems later.The solution method for kinematics problems converts the known conditions expressed in the form of triples into readable analysis processes and solution steps.Firstly,by combining the characteristics of the kinematics problem with the predicate logic reasoning,the storage mode of the rules for solving the kinematics problem is designed.The forward chain inference method is then used to solve the kinematics problem.The problem to be solved is regarded as the target of reasoning,and the physical knowledge and known conditions are regarded as rules and existing facts respectively,and then the final answer is solved through reasoning.Finally,three levels of understanding of the problem are used to evaluate the performance of the system,and verification experiments are carried out on an authoritative source kinematic problem dataset.The results show that the algorithm successfully solves92.59% of the kinematic problems,which is higher than the baseline algorithm. |
