Text confrontation attack method based on style migration and whale optimization algorithm hybridization

The invention discloses a text confrontation attack method based on style migration and whale optimization algorithm hybridization, which belongs to the field of text confrontation attack, and comprises the following steps: constructing a three-stage module and an improved style migration module; generating a text adversarial sample through an adversarial sample generation model based on a sememe, and inputting the text adversarial sample into an improved style migration module and a three-stage module for attack; and if the improved style migration attack fails, entering the attack of the three-stage module. The problems that an existing single model is poor in attack success rate performance, an effective confrontation sample is difficult to search, and semantics is prone to being deviated are solved; the fusion model is used for attacking in the text adversarial attack field for the first time, and the attack success rate is improved. 本发明公开了一种基于风格迁移与鲸鱼优化算法杂交的文本对抗攻击方法，属于文本对抗攻击领域，包括以下步骤：构建三阶段模块和改进风格迁移模块；通过基于