Explicit and implicit variational enhancement combined multi-document literature abstract method and system

The invention discloses a multi-document literature abstract method and system combining explicit and implicit variational enhancement, and the method comprises the steps: capturing explicit statement topic representation of an input document through a neural topic model, and fusing the explicit statement topic representation with an initial abstract statement or an output previous abstract statement to obtain explicit fusion features; then, the hidden variable model is used for capturing to obtain hidden upper-layer features, and a sentence decoder is used for predicting sentence planning of a next sentence according to the hidden upper-layer features; and inputting the initial abstract statement or the output previous abstract statement into a word decoder, and combining the sentence plan of the next sentence to obtain the next word until the sentence plan of the next sentence is a terminator. According to the abstract generation method and device, abstract generation of sentence planning control can be ach