Research On Automatic Scoring Method For Subjective Questions In Transportation Field Based On Multi-feature Fusion

With the continuous development of modern information technology,educational informatization has entered a new stage,presenting trends towards intelligence,personalization,openness,and collaboration.The deep integration of artificial intelligence and education has become an important means of enhancing the level and quality of educational informatization,heralding the dawn of the era of intelligent education in the field.Automated scoring,in particular,is a crucial aspect of intelligent education,as it not only greatly reduces the heavy workload of teachers,but also avoids the phenomenon of subjective grading errors caused by teachers,thus ensuring the fairness of exams.Students can also use automated scoring for self-evaluation anytime and anywhere,to better consolidate their knowledge.Moreover,automated scoring provides a guarantee and convenience for large-scale online teaching and exams.In the field of automated scoring,the subjective automated scoring of the transportation industry is a practical and theoretical issue with strong research and application value,involving theories and knowledge of transportation planning,transportation information technology,and transportation safety,among other areas.Currently,automated scoring methods for objective questions have matured and have been extensively researched and applied.However,in the realm of subjective questions,automated scoring methods still face many challenges and issues that require urgent solutions.In the field of transportation,existing automated scoring methods for subjective questions are inadequate in capturing the deep semantic meaning of student answer texts,while also facing challenges such as diverse syntactic structures and insufficient knowledge of specific domain expertise.Therefore,this study proposes a transportation-specific subjective question automated scoring method based on multi-feature fusion.It mainly includes the following aspects:(1)By using the BERT model to dynamically encode the answer text,semantic features of the answer text can be obtained,effectively improving the accuracy of semantic analysis,thus better understanding student expression.This method reduces semantic analysis errors and solves the problem of semantic diversity.In response to the diverse syntactic structures of student answers,the syntax features of answer text are constructed by using a syntax dependency graph.This method solves the problem of significant differences in sentence structure,language style,and text length that may exist in student answers.In response to the specific knowledge in the field of transportation that may be present in student answers,domain-specific knowledge features of the answer text are constructed by using an entity graph.This method solves the problem of insufficient understanding of domain-specific terminology in this field.(2)Propose a comprehensive method for automatic scoring of subjective questions that integrates semantic features,syntactic features,and domain-specific knowledge features.Firstly,utilize semantic features to supplement missing semantic information in the syntactic dependency graph and entity graph,respectively,to obtain the extended syntactic dependency graph and extended entity graph.Secondly,propose a feature fusion formula that dynamically fuses the extended syntactic dependency graph and extended entity graph into an overall feature fusion graph.Next,encode and capture the global information of the answer text through a multi-layer graph convolutional neural network on the overall feature fusion graph.Then,add a Co-attention layer to the graph convolutional neural network to obtain the local information of the answer text.Finally,combine the global and local information of the answer text to complete automatic scoring.(3)Based on the existing multi-feature fusion automatic scoring method in the transportation field for subjective questions,this study developed an automatic scoring system for the transportation field and validated its effectiveness through experiments.