Sliding window error correction-based attack resisting text enhancement method and storage device

Abstract

The invention relates to the technical field of natural language processing, in particular to a sliding window error correction-based anti-attack text enhancement method and storage equipment. The method for enhancing the attack resisting text based on sliding window error correction comprises the following steps: acquiring an original text to be processed; performing sliding window processing on the original text to generate a plurality of original clauses; performing text error correction on the original clauses subjected to sliding window processing through a text error correction tool to generate corrected clauses; and performing cross combination on the original clauses and the corrected clauses to generate a new text. According to the method, the sliding window processing and the text error correction tool are combined to play a role in double data amplification, correction and amplification can be carried out only at the word or word level line of the clause in combination with the error correction tool, the semantics is hardly influenced, and the robustness and generalization ability of the model are effectively improved under the condition that the semantics is not influenced.

Description

technical field [0001] The invention relates to the technical field of natural language processing, in particular to a method and a storage device for text enhancement against attacks based on sliding window error correction. Background technique [0002] The application number is CN111507114A (published on August 07, 2020) which discloses a method for enhancing spoken text based on reverse translation. The method includes: training a label prediction model through the spoken text with the first set of labels; translating the spoken text with the first set of labels into other languages, and then reverse-translating to generate multiple enhanced spoken texts; The prediction model performs label prediction on multiple enhanced spoken texts, and determines the second set of labels corresponding to each of the multiple enhanced spoken texts; respectively judges the labels in the first set of labels, and in the second set of labels corresponding to the multiple enhanced spoken t...

AI Technical Summary

Problems solved by technology

[0015] To this end, it is necessary to provide a text enhancement method based on sliding window error correction against attacks to solve the technical problems of semantic loss, poor robustness, and poor generalization ability caused by the introduction of large noise in existing text enhancement methods. The specific technical solution as follows:

Images