Research On Information Extraction Method Of Junior Middle School Mathematics Exercises Based On Deep Learning

In recent years,with the rapid development of artificial intelligence technology,a variety of intelligent products began to be applied in various fields,replacing people to complete a lot of complex and tedious work,greatly reducing people's burden and improving efficiency.In the field of education,how to use intelligent technology to replace or partially replace human teachers for one-to-one tutoring of students has always been the goal of people's pursuit.At present,students encounter difficulties in solving problems after class,and the common method to solve problems with technology is to take pictures to search for problems.This method allows students to quickly get answers,or video explanations of the topic.In order to promote the development of students' thinking,some products adopt the way of step-by-step tutoring,which can meet students' personalized learning needs and improve learning efficiency in certain procedures.However,these methods have many disadvantages,such as the need to invest a lot of manpower,the platform can not realize the interaction with students,and can not answer students' personalized questions like real teachers' one-to-one tutoring.If the system to achieve real "intelligent",automatic extraction of key information in the topic,is the first step.In this paper,deep learning and natural language processing technologies are used to study the junior middle school math problems of Beijing Normal University edition,aiming to achieve automatic extraction of problem solving information in math exercises.The main research contents include the following:(1)the key information of junior high school math exercises.Firstly,the characteristics and concepts of mathematical language are analyzed,and then mathematical entities,mathematical events and the ellipsis of reference in mathematics are studied.(2)Named entity recognition for mathematical exercises.The deep learning method is adopted,and the bidirectional short and long time memory network and conditional random field model are used.After manual annotation and data pretreatment,the training model with high accuracy is obtained,and the corresponding entities and attributes can be output after the input of mathematical questions.(3)Event extraction of mathematical exercises.According to the basic steps of event extraction,mathematical questions are extracted as an event.Combined with the deep learning model and rule-based method,the topic clauses are identified,event types are identified based on trigger words,event elements are extracted,and the key information of solving each question is obtained.(4)Anaphora resolution of mathematical exercises.In this paper,a complete and easy to operate rule is formulated by analyzing various reference relations of mathematical problems.Based on the rule,the reference phenomena in the problems can be eliminated and the ambiguity can be eliminated in the process of understanding the meaning of the problems.The experimental results show that the named entity recognition method,event extraction method and reference resolution method can better achieve the key information extraction of mathematical exercises.