Test paper structure analysis method based on span information and local attention

The invention discloses a test paper structure analysis method based on span information and local attention, and the method comprises the following steps: obtaining an electronic test paper, and preprocessing multivariate data in the electronic test paper into a unified branch structural text; the method comprises the following steps: firstly, extracting semantic features of each row of test paper through a large-scale pre-training model, and modeling context information; then utilizing supervision attention based on span information and an auxiliary task based on span classification to explicitly model segmentation structure information of the test paper and feature information based on span, and splitting the test paper into different types of big question segments and corresponding types through a classifier; and further processing each obtained big question segment, splitting the big question segments into small questions by using a classifier and various previously extracted feature information, and ret