Novel network media platform variant comment adversarial text generation method

Abstract

The invention provides a novel network media platform variant comment adversarial text generation method, which comprises the following steps of: on the basis of summarizing variant text variant rules commonly used by a novel network media platform, firstly, extracting feature words from classified annotated texts; carrying out variant vocabulary generation based on various rules on the feature words, and carrying out variant text generation based on variant rules on the basis of the variant vocabulary generation; then training the annotated text through a word2vec word vector method to obtain word vectors of all vocabularies, obtaining a similar word list of all vocabularies according to the word vectors, and achieving variant text generation based on neural network word vectors; and finally, achieving a variant text generation method combining variant rules and word vector similar words through a probability randomization method. According to the method, massive variant texts in different forms can be generated, conventional text filtering can be resisted, and high fidelity is achieved.

Description

Technical field [0001] The present invention relates to a natural language processing review text generation technology, and is specifically a new type of network media platform variant comment against text generation method. Background [0002] New network media platforms, such as shake, fast hand, Netease cloud music, etc. have hundreds of millions of account users, in which some bad users are used to avoid review of spam comments produced by variant methods such as homogenesis. These variant spam comments with negative emotional or poor metaphors have seriously polluted network environments, causing negative impact on platform users, analyzing and correctly identifying these variant spam comments. It is of great significance for platform health development. [0003] The existing variant garbage text classification method mainly includes two categories, one is based on variant word identification and normalization, by extracting variant word characteristics identifying text var...

AI Technical Summary

Problems solved by technology

[0005] Aiming at the problem that the variant text classification accuracy of the variant text classification method based on the deep neural network is not high due to insufficient training data, the present invention analyzes the variant text rules and utilizes feature word extraction, word sequence randomization, word Vector and text generation techniques enable different forms of variant adversarial text generation

Images