Automatic generation method for reading understanding problem

The invention provides an automatic generation method for a reading understanding question, and belongs to the field of question generation of natural language processing, and the method comprises the steps: carrying out the data filtering of an RACE data set, carrying out the marking of answer words through an answer marking method, and reconstructing a QG data set RACE4QG; constructing an end-to-end QG model, wherein a GRU of an encoder in the QG model takes splicing of vectors generated by word embedding, answer marking and GAT as input; the hidden state of the encoder obtains a final channel-answer representation through a gating self-attention mechanism, and the final channel-answer representation is input into a decoder; and decoding the problem generation model. Aiming at the defects that the period is too long and time and labor are consumed when an English teacher manually generates a reading understanding problem, a deep neural network method is adopted to automatically generate the reading understa