Research On Automatic Scoring Technology For Chinese Composition

At this stage,the evaluation mode of Chinese composition is still based on manual evaluation,but manual evaluation is not only time-consuming and labor-intensive,but also has the problem of low scoring reliability due to the inevitable subjectivity of scoring.Automatic composition scoring can use computer technology to score composition objectively and efficiently.The automatic composition scoring system based on English has been widely used,but the automatic scoring system for Chinese composition is still in the exploratory stage.With the improvement of natural language processing technology and computing power,the realization of automatic scoring of Chinese composition has attracted more and more attention from researchers.The automatic scoring of Chinese composition has a wide range of application scenarios.The uniqueness and complexity of Chinese language makes the realization of Chinese automatic scoring full of challenges.This paper investigates the automatic scoring technology of domestic and foreign compositions,using machine learning algorithms and deep neural network algorithms to explore automatic scoring technologies for Chinese compositions based on shallow and deep features,and further improves the scoring effect through integrated learning.The main research contents are as follows:The composition scoring based on shallow features,refer to existing scoring scales and related research on scale design for various types of essays,and combine HSK composition features to design shallow features that reflect the accuracy,fluency,richness,and completeness of the composition.Naive Bayes,K-nearest neighbor,support vector machine and random forest four machine learning algorithms are used to train the scoring model and compare the scoring effects in different situations.The composition scoring based on deep features,after vectorized composition by Word2 Vec,the text application algorithm of convolutional neural network is used to train the scoring model.Improve the model effect through hyperparameter optimization,and design experiments to compare the different performance of convolutional neural networks in different situations to understand the applicability of convolutional neural networks in HSK composition quality classification.The composition scoring based on integrated learning,combined with the prediction results of the scoring model of shallow and deep features,and the majority voting strategy of integrated learning,effectively improve the scoring effect of Chinese composition automatic scoring.Effectively weaken the shortcomings of a single model,reduce the risk of its misjudgment,so as to achieve a more scientific,comprehensive and objective Chinese composition scoring.