Research And Implementation Of Automatic Stop-reasoning Technology In Automatic Problem-solving System Of Elementary Mathematics

The rapid development of artificial intelligence technology has led the technological transformation of all walks of life.Hot issues in the field of intelligent education represented by automatic problem-solving technology and automatic assessment technology have attracted more and more researchers’ interest.The evolution of artificial intelligence from perceptual intelligence to cognitive intelligence requires machines to have the ability of logical reasoning.Solving complex mathematical problems is a process of logical reasoning.Allowing machines to automatically solve mathematical problems is an important criterion for measuring whether a machine has logical reasoning ability.In the process of problem-solving reasoning,it is an important task to automatically identify the solving objects in the problem and automatically identify the correct solution result.Based on the above background,this thesis studies and realizes the automatic stop-reasoning technology in the automatic problem-solving system of elementary mathematics.The main research contents are as follows:1.A method of expressing the solving objects in elementary mathematics texts.Identify mathematical entities and extract the relationship between mathematical entities from the text,then represent the text as <entity,relationship,entity> triplets.Based on triplets,a labeling method for solving objects is designed.2.A method of recognizing solving objects in elementary mathematics texts.An automatic identification method for the solving object in the problem triplets is designed.The implementation of this method combines syntactic analysis and neural network models.In the case of no manual intervention,the method provides solution targets for the reasoning system.Solving objects guide the direction of the reasoning process and avoid the problem of branch explosion.3.Design of stop-reasoning strategies in automatic problem-solving system.Referring to the problem-solving methods of elementary mathematics,five stopping reasoning strategies are designed considering the existing structure of the problem-solving system.It can effectively judge the stop-reasoning state in the process of solving problems,and further improve the correctness of the problem-solving system.4.A split-and-match method of stop-reasoning nodes is proposed.Based on the identification method of the solving objects and the characteristics of reasoning intermediate results,the concept of stop-reasoning nodes is proposed.In the reasoning process,the splitting of stop-reasoning nodes and the matching method of stop-reasoning nodes with new knowledge generated by reasoning are designed.Make the reasoning system stop reasoning normally in time and avoid invalid reasoning process.Finally,the above method is successfully applied to the automatic problem-solving system of elementary mathematics.Realized the solving objects recognition module and stop-reasoning module in the system.Without human intervention,the system can mark solving objects in the text of math problems and realizes automatic stop reasoning during the reasoning process.Prepare for the system to output human-readable solution steps.At the same time,different stopping reasoning strategies are designed to support singlebranch single-solving-object reasoning and multi-branch multi-solving-object reasoning.The system is endowed the ability to solve classified discussions and multi-solution problems.On the constructed test set,the problem-solving accuracy rate reaches 82.1%.