Research On Chinese Automatic Scoring Technology Based On Short Text Similarity And Smoothness

In recent years,the rapid development of artificial intelligence promotes the online education system to develop in the direction of information.As the main part of online education system,Chinese automatic marking technology has been concerned and studied by more and more people.At present,the automatic marking technology of objective questions such as multiple choice questions and judgment questions has been relatively improved.However,the automatic marking of subjective questions such as short answers is mainly due to factors such as the language complexity of the Chinese text.The workload of manual marking is large,and the efficiency is low,so it is hard to avoid the deviation of grading in some cases.Therefore,the research on the automatic marking technology of subjective questions has important practical significance.This paper mainly studies the two important scoring features of short text similarity and text fluency.The specific work is as follows:In this paper,a short text similarity algorithm based on attention and syntax is proposed.In order to solve the problem that the different importance of different parts of the text can not be fully utilized and the syntactic information is seldom considered in the calculation of short text similarity,which leads to low accuracy,we comprehensively consider the factors that affect the short text similarity,improve the semantic similarity and syntactic structure similarity respectively.In the part of semantic similarity,a mixed attention structure system is designed to accurately extract interactive information and highlight the importance of different granularity;in the part of syntactic structure,a new graph-based dependency analysis method is proposed to accurately analyze the dependency of the text,and construct the relation matrix,calculate the structural similarity;finally,the weighted fusion of the two.The experimental results show that this method is better than other text similarity methods.In this paper,a text smoothness algorithm based on neural network is proposed.This paper mainly improves the modeling of sentence embedding representation and inter-sentence smoothness.In order to solve the problem of inaccurate sentence representation caused by inaccurate word segmentation,a word joint vector construction algorithm based on dictionary is proposed to obtain the joint vector which is not affected by word segmentation.Then a CNN network with improved pooling strategy is proposed to extract features,and position information and intensity information are added to the sentence representation to obtain accurate sentence embedding representation.In order to solve the problem that only the semantic similarity between adjacent sentences is concerned and the rhetorical relationship between sentences is ignored in current smoothness modeling,it is proposed to unify sentences and use multiple attention mechanism to model the position information of adjacent sentences in each unit.Bilinear tensor is used to simulate the rhetorical relationship between sentences to improve the accuracy of text smoothness modeling.The experimental results show that the algorithm achieves good results on the data set.In this paper,on the basis of the improved short text similarity algorithm and text smoothness algorithm,the two are combined by binary linear regression to construct an automatic scoring model for subjective questions.A man-machine interactive page with automatic scoring is designed for this model,and the function of the automatic scoring model is visualized.A comparative experiment is carried out on the financial and commercial examination papers of vocational secondary schools in Shanxi Province,and the results show that the automatic scoring model basically meets the requirements of automatic marking,and has a certain accuracy and effectiveness.