Knowledge enhancement-based text generation model and training method thereof

The invention provides a text generation model based on knowledge enhancement, which is used for generating a text sequence related to an input text sequence according to the input text sequence, the model comprises a coding module and a decoding module, the coding module comprises a coding unit used for coding each word in the input text sequence into a hidden vector, obtaining a first semantic vector of the word according to each word hidden vector; the knowledge graph attention unit is used for obtaining a knowledge graph vector corresponding to each word based on the knowledge graph, and splicing the knowledge graph vector of each word and the first semantic vector of the word to obtain a second semantic vector of each word; the first variation sampling unit is used for performing variation sampling on the second semantic vector to obtain a first hidden space vector; the regression sampling transformation unit is used for carrying out autoregression sampling on the first hidden space vector to obtain a se