An Automatic Scoring Method For Subjective Questions Using Semantic Technology And LSTM

In recent years,with the rapid development of science and technology,artificial intelligence has been increasingly applied to our daily lives.Computer-assisted instruction has been favored by more and more people because of its convenience,speed,and intelligence.There are many problems with the traditional scoring method.On the one hand,it causes a great workload and reduces the time for preparing lessons for teachers;on the other hand,the manual scoring of subjective questions has strong subjectivity and unfairness.At present,online learning,online examinations and other related technologies are becoming more and more common.How to release people from the fatigue of scoring and critique of examination papers has become an urgent issue.Owing to the joint efforts of experts and scholars at home and abroad,it has been able to accurately achieve automatic scoring of objective questions such as single-choice questions,multichoice questions,judgment questions,fill-in-the-blank questions and so on.For the subjective questions(such as short answer questions,noun explanation questions,essay questions and so on)expressed in natural language for the answer,in the case of intelligent scoring,there are differences in the way of thinking,the degree of understanding,and the habits of expression,resulting in differences in the content of subjective answers.When teachers scoring subjective questions,they are often influenced by many subjective factors,which make the marking is unfair.Each of the existed subjective questions automatic scoring methods have different application backgrounds,and the final calculation result standards are not the same.Each method is not very versatile and has different degrees of limitations.In order to improve the adaptability and practicability of the automatic scoring methods for subjective questions based on syntactic analysis,dependencies analysis and ontology,this dissertation makes full use of the semantic dependencies between sentence components,which is independent of their physical position,the inclusiveness of the answers and the universality of Word Net,the uniqueness of LSTM neural network that can remember the long-term information,and proposes an automatic scoring method for subjective questions using semantic technology and LSTM.The main work of this dissertation includes the following points:(1)This dissertation construct the domain ontology of the subject “Computer Network” to expand the knowledge base of the common ontology Word Net.The constructed domain ontology is mainly used for the calculation of word similarity,which makes up for the lack of proper nouns in the common ontology Word Net during the automatic scoring of subjective questions,and increases the accuracy of word similarity calculation.(2)This dissertation calculate the word similarity between adjectives or adverbs through similar-to or antonym relationships between adjectives and adverbs,and increase the accuracy of word similarity calculations.And this dissertation use the dual similarity calculation method combining the domain ontology of the subject “Computer Network” and the common ontology Word Net to calculate the word similarity.The maximum similarity calculated based on the Word Net and the domain ontology is used as the similarity value of the student's answer and the standard answer,which guarantees the efficiency and accuracy of the automatic scoring of subjective questions.(3)This dissertation built an LSTM neural network classification model to achieve automatic synchronous double classification(title type classification and question type classification)of the subjective question corpus.It provides the early guarantee for the realization of different question types using different subjective question automatic scoring methods,and provides the possibility to improve the accuracy of automatic scoring of subjective questions.(4)This dissertation use syntax analysis and dependencies analysis to analyze questions and answers(standard answers and student's answers).Through the dependencies analysis of title to get the components of interrogative words in questions.Through the dependencies analysis of the answer to get the core semantics,non-core semantics,the head words and the negative tone of the answers(standard answers and student's answers).Different question types use different subjective question scoring methods to achieve adaptive scoring of different answer methods of the same standard answer,which further improves the accuracy and practicability of the subjective question scoring system,and makes the intelligent scoring of the subjective questions more flexible and efficient.