An automatic grading method and system for a picture-based composition

Abstract

The invention discloses an automatic grading method and system for drawing and writing type composition. The method comprises obtaining the n-gram feature corresponding to the composition text, obtaining the sentence representation corresponding to the composition text, obtaining the hidden layer output matrix corresponding to the sentence representation, obtaining the feature vector matrix corresponding to the picture, and calculating the The hidden layer outputs the similarity matrix corresponding to the feature vector matrix, performs the first attention operation and the second attention operation, calculates the mutual information matrix, and inputs the mutual information matrix to the fully connected layer, and outputs the scoring score and other steps. The present invention fuses the text semantic information in the output matrix of the hidden layer with the visual semantic information in the feature vector matrix by implementing a co-attention mechanism, and can realize automatic scoring for writing-type composition based on looking at pictures. The invention is widely used in the technical field of automatic scoring of writing-type compositions by looking at pictures.

Description

technical field [0001] The invention relates to the technical fields of image recognition and natural language processing, in particular to an automatic scoring method and system for writing composition by looking at pictures. Background technique [0002] Picture-based writing refers to the process in which candidates write according to the pictures given in the topic. One of the scoring criteria for picture-based writing is the degree of agreement between the content reflected in the composition text written by the test-taker and the content reflected in the topic picture , that is, the more the composition text can fully and accurately reflect the content of the title picture, the higher the score of the composition text will be. [0003] Modern examinations tend to be scaled and standardized, which makes manual marking more and more difficult. Due to the strong subjectivity of manual marking, it is difficult to maintain stable marking standards and high marking efficien...

AI Technical Summary

Problems solved by technology

The principle of computer automatic marking technology based on neural network is to convert the words in the composition text into word vectors and input them to the convolutional neural network or recurrent neural network for sentence coding and article coding, and finally obtain the composition representation containing the high-level abstract semantic features of the composition. However, since this technique does not capture the semantic information in the topic picture, it can only be applied to topic composition and other topic types, and it is difficult to meet the scoring requirements of picture-based composition

Images