Research And Implementation Of Automatic Scoring System For Subjective Questions Based On Text Similarity Calculation

In the current society,with the continuous expansion of the online and offline education market,the demand for all kinds of examinations is increasing.However,the traditional manual marking method can not meet the needs of modern society because of its high cost,timeconsuming and slow statistics.Although the technology of automatic scoring for objective item is already mature and widely used.However,because of the complexity of the Chinese language,it still has many disadvantages.For example,the impact of semantic and word order issues on the scoring accuracy is not considered in this technology.At the meantime,for major issues such the visibility of practicability and visibility of the model scoring process,there is not numerous automatic scoring system for Chinese objective been applied yet.At last,to solute those problems above,this paper will proposes two technical routes to support the algorithm of automatic scoring system for subjective questions and proved it by experimental data of logistics students’ exam paper.1.Automatic scoring model of subjective questions based on text similarity.In order to solve the scoring problem of subjective questions in logistics specialty,this paper adopts Siamese Network model and named entity recognition method,further through the model integration to mark of subjective questions.The deep learning model based on Siamese Network can ensure the accuracy of the results.At the same time,named entity recognition method can extract the student answer and the standard answer entity sequence matching.The named entity recognition method can also effectively improve the computational efficiency of the model and the efficiency of long text matching.2.Automatic scoring model for subjective questions based on knowledge graph.In order to solve the problem that the scoring model based on text similarity is not effective in scoring professional noun interpretation questions,this paper proposes an automatic scoring model for subjective questions based on knowledge graph.A total of 5314 pairs of triples related to logistics major are constructed by using the corpus of logistics major,and trained by knowledge embedding model.Knowledge graph has a good representation and integration of semantics,and can quickly and effectively obtain the extended relationship between knowledge.Therefore,it is suitable for questions such as noun explanation of logistics majors.In this paper,the subjective question scoring models based on two different methods are trained and their results are compared respectively.For the model based on text similarity,the loss value of the trained scoring point identification model converges at about 0.9,and the accuracy of the model is 80.54%.The accuracy of the trained text similarity matching model is86.99%.After data fusion,the scoring time of a single use case of the model is less than 0.8s,and the MSE was 0.85 for short answer tests and 1.61 for noun interpretation questions tests.For the embedded model based on knowledge graph,the MRR value after training is 0.3582,Hits@10 is 0.3685 the MSE of the noun interpretation test data is 0.45,which is significantly better than the subjective test score model based on text similarity model.Based on algorithm above,this paper build an automatic scoring system of subjective questions,and provides corresponding operating interfaces for different user groups.