251 / 2018-08-31 08:32:16

Research on automatic composition scoring based on latent semantic analysis and linguistic features

Automatic scoring of composition; latent semantic analysis; BP neural network

全体主题 > 人工智能技术在语言测评中的应用

Automated composition scoring can not only solve the problem of efficiency, but also give full play to the advantages of process approach, autonomous learning and self-construction to stimulate students'interest in writing. Based on the existing HSK composition corpus, this study firstly uses natural language processing method and statistical linguistic features, then constructs a data matrix composed of words and documents according to latent semantic analysis method, carries out matrix singular value decomposition, calculates cosine distance similarity in training composition documents, and then obtains training. The functional relation of sample level of composition is used to obtain the score level of the composition to be tested; then the score level is taken as an important variable in the input layer of BP neural network, and a new feature matrix is constructed by combining the surface features of language such as vocabulary usage and sentence composition, and then the neural network is trained; finally, the basis is obtained. The composition scores of BP neural network. It is verified that the accuracy of scoring based on BP neural network will be improved if the high phase variables are added.