Discovering Hidden Entities Of Mathematical Word Problems Based On Deep Learning

The automatic answering of mathematics problems in primary and secondary schools has received extensive attention from many researchers for many years,because it is an interesting question.It’s worth the challenge.However,with the rapid development of machine learning and pattern recognition technologies in recent years,the combination of artificial intelligence-related technologies and machine answers to math problems in primary and secondary schools has become a new research hotspot.Specifically,there are two important steps in machine answering,namely the question understanding process and the question answering process.The research goal of this paper is the intelligent solution of mathematical application problems in primary and secondary schools.At present,researchers have solved the directly stated mathematical application problems through the S2 syntactic and semantic hybrid model,and proposed methods for extracting the implicit quantitative relations in the mathematical application problems.However,S2 model still has some mathematical problems that cannot be solved.For example,when a sentence contains multiple semantic models,the problem cannot be solved.For more complex problems,when there are multiple intermediate variables in the problem-solving process.the effect of the model will be unsatisfactory.Therefore,this paper studies the process of solving math problems contains multiple intermediate variables.On this basis,the concept of implicit entity is proposed and transformed into the task of entity relationship extraction.The research work based on deep learning mainly includes the following contents:First,by disassembling the solving steps of mathematical application problems and connecting multiple equations,the implicit entities in mathematical application problems are defined.Through its analysis,it can be found that the discovery of hidden entities is essentially the identification of quantitative entities in mathematical application problems and the extraction of corresponding relationships.On this basis,the construction of the corpus of hidden entities and the discovery of hidden entities are carried out.Secondly,use the Chinese word segmentation tool to segment the original mathematics application problem text,and label the quantitative phrases and the solution targets.According to the corresponding problem-solving equations and problem-type formulas,the equations are disassembled,and the quantitative phrase entities are connected to realize the labeling of the implicit entities,and finally the construction of the mathematical application problem corpus is completed.Finally,according to the relevant literature in the field of relation extraction in recent years,the author built two relation extraction models based on deep learning,and pretrained on the mathematical application problem corpus based on the established models.In addition,this research uses the trained model to conduct hidden entity discovery experiments and multiple hidden entity discovery experiments to generate problem-solving steps.number of entities in the question,thereby forming the answering step.