Method, device and equipment for identifying risk content

Abstract

The embodiments of this specification provide a method, device, and equipment for identifying risky content. The method includes: for the target text to be identified, identifying whether the target text contains a risk text segment, and the pinyin corresponding to the risk text segment belongs to the set risk pinyin; if the target text contains the risk text segment, generating A feature sequence corresponding to the target text, the feature sequence including: the text features contained in the target text and the pinyin features corresponding to the risk text fragment; based on the feature sequence and a model obtained through machine learning algorithm training , to determine whether the target text belongs to risk content, the input of the model is a feature vector corresponding to each feature in the feature sequence, and the output of the model represents the possibility that the target text belongs to risk content.

Description

technical field [0001] One or more embodiments of this specification relate to the technical field of machine learning, and in particular to a method, device, and device for identifying risky content. Background technique [0002] At present, in order to maintain the purity of the Internet environment, it is necessary to identify and deal with undesirable risky content (such as tyranny, terrorism, pornography, gambling and drugs, and spam advertisements, etc.). Generally, an identification engine can be built for different risk content to identify risk content. However, in order to prevent the published content from being identified as risky content, information publishers create more and more variant information, which makes it difficult to identify risky content. For example, use homonyms to replace sensitive words in sentences that originally belonged to risk content, for example: "I lost 41,000, dog, fart, chicken, gold, cut the meat", among them, use "chicken gold" ins...

AI Technical Summary

Problems solved by technology

However, in order to prevent the published content from being identified as risky content, information publishers create more and more variant information, which makes it difficult to identify risky content

Images