Fully-generated knowledge question-answer pair generation method based on pre-training model

A fully-generated knowledge question-answer pair generation method based on a pre-training model comprises the following steps: selecting an original data set, and processing the original data set into result; text, question and answer gt; a format; learning the high-level semantic representation of each word in the text and the final output representation of the question and the answer through a pre-training model; the output expression of the answer and the learned text high-level semantic expression are combined, words can be copied from a source text by means of a pointer generation network, and finally the final answer is generated through a generator; after the answers are generated, the generated information is fused into output representation of the questions through a multi-head attention mechanism guided by the answers, and finally the questions are generated through a generator. According to the method, semantic compatibility of answer and question generation is considered, cross-task communication