Unsupervised suggestion abstract generation method and system based on antagonism framework

The invention discloses an unsupervised opinion abstract generation method and system based on an antagonism framework. The method comprises the following steps: firstly, constructing a data set, selecting to-be-tested data in a database, preprocessing the data, dividing the data for verification and test, and constructing an unsupervised opinion abstract data set; constructing an M2A model, selecting two abstract generation models as abstract generators to embody the model independence of the M2A, and constructing an abstract discriminator of the model based on a natural language reasoning method; secondly, firstly training an abstract discriminator, then integrally training an M2A model, constructing a loss function, and constructing an optimizer; and finally, evaluating the M2A model through the evaluation index. According to the method, the category accuracy and emotion accuracy of the generated opinion abstract are remarkably improved, and the working efficiency of abstract generation is greatly improved