Multi-language harmful information feature intelligent mining method based on deep learning

Abstract

The invention discloses a multi-language harmful information feature intelligent mining method based on deep learning. The method comprises the steps of marking harmful and harmless information textsof all languages and all classes; selecting candidate words from each category of words of each language by using an RNSW method and establishing a one-hot code code; inputting the sample data into aCNN neural network model for training to obtain a score that each word belongs to the harmful category of the language, and taking the score as a weight; and screening the harmful information featuresselected by machine learning by using a genetic algorithm to form final harmful information features and weights. The invention provides a language-independent text dimension reduction representationRNSW method, effectively reduces the number of parameters of model training, accelerates the training speed, and improves the accuracy of model recognition; and secondly, the intelligent mining of harmful information features is realized by adopting a deep learning method, and the harmful information features are screened through a genetic algorithm, so that the interpretability of harmful information identification is better.

Description

Technical field: [0001] The present invention relates to text analysis technology in the Internet field, in particular to a harmful text recognition method, which is an intelligent mining method for multilingual harmful information features based on deep learning. Background technique: [0002] There are two commonly used methods for identifying harmful information, one is based on keyword and rule matching, and the other is based on machine learning. Based on the method of matching keywords and rules, it is necessary to manually edit the lexicon of harmful words. Sometimes the rules are complex enough to achieve better results. However, harmful words and new words emerge one after another on the Internet, and the update cycle is short. Maintaining the lexicon and designing new rules It costs a lot of money. The method based on machine learning is gradually adopted in recent years. The advantage of this method is that it does not require technicians to have in-depth domain ...

AI Technical Summary

Problems solved by technology

This method requires images in web pages as auxiliary information for identification, but most web texts do not have pictures, and some pictures may have nothing to do with the text, so this method is not easy to use on large-scale text data

Images