Automatic Classification Of Question Types Based On Multi-feature And Its Application In Intelligent Item Bank

In recent years,with the growth of data and the development of computing power,the theory and technology of artificial intelligence has made breakthroughs and received extensive attention in various fields.In the field of education,artificial intelligence technology can be used to optimize teaching procedure,mining educational data,customizing personalized teaching scheme,etc.,which has positive significance for the development of education.Intelligent item bank is an important component of educational intelligence,and question types classification servers as the basis of intelligent item bank tasks,such as papers automatic generation,personalized recommendation and unit training.The aim of question types classification is to learn the mapping between the question text and the category(question type)based on given problem text and question type,so that the type of question can be automatically judged based on the input,which can be regarded as a text classification tasks of vertical field.Considering the significant difference of language styles between subjects,this thesis chooses to study the automatic classification of elementary mathematical problems.First,the theory and method of text classification in common domain are introduced,then several key issues are studied,including mathematical text preprocessing,feature extraction and feature representation,and the design of classification model with multi-features,etc.,as follows:1.Feature extraction of elementary mathematical text.Mathematical texts have the commonality of ordinary natural language texts,deep neural networks have powerful representation ability for texts.Therefore,this thesis uses neural networks to extract text-level features automatically,to reduce feature engineering and learn deep representations.On the other hand,in order to extract the mathematical logic features,such as mathematical transformation,the application of theorem,etc.,automatic knowledge point annotation technology based on automatic reasoning is used to extract the mathematical features of a given question.2.Learning feature representations based on word embedding.The text feature sequence and the knowledge point feature sequence are transformed into word embedding which can be processed by the neural network.On this basis,a suitable deep neural network structure is designed to learn representation of question,and a deep neural network classification model with multiple features is proposed.Based on the above ideas,an automatic classification system of mathematical problems is designed and implemented,which solves the preprocessing methods for grammar in mathematics,the improvement of general natural language processing methods in the field of mathematical texts,the implementation of proposed model and communication between modules,etc.The process from input to output is automated.Finally,the model is trained on a sample composed of 64950 junior high school mathematical problems,and several comparative experiments are designed to verify the effectiveness of the proposed method.The experimental results show that the multi-feature automatic classification system for elementary mathematical problems designed in this paper can effectively synthesize the textual and logical features of mathematical texts,improve the classification accuracy,meet the requirements of intelligent item bank fairly,and has good expansibility,which has a certain practical value.