Network information security detection method and device

Abstract

The invention discloses a network information security detection method and device, and the method comprises the steps: splicing characters in a scene, and inputting the spliced characters into a character encoder for security detection; for the pictures in the scene, extracting related features violating concepts and corresponding degrees for each picture through a pre-trained single-picture machine learning classification model; selecting one or more pictures with the highest degree from all the pictures according to the related characteristics violating the concept and the corresponding degrees, and inputting the selected pictures into a picture encoder for security detection; and splicing the output layers of the character encoder and the picture encoder to obtain a multi-modal fusion model. The system comprises a character splicing unit, a character encoder, a picture feature extraction unit, a picture encoder and a multi-modal fusion unit. According to the method and the device, a mode of multi-modal machine learning and feature extraction is used, so that the learning model can still perform successful detection in a complex scene mixed by various media contents.

Description

technical field [0001] The invention relates to the technical field of network security, in particular to a network information security detection method and device. Background technique [0002] In recent years, people have paid more and more attention to network information security. In the direction of Integrity, whether it is because of regional policy requirements or increasing user needs, major domestic and foreign companies such as Google, Facebook, and Tencent have invested heavily in the establishment of trust-and-safety / integrity / Tencent You Figure 1 A series of departments are used to protect and create a safe and healthy network / APP environment. [0003] Generally speaking, network information security is divided into two parts, content (content) and user (user). The former focuses on whether the information published / uploaded by users in the APP / network conforms to the network / community system, and the latter focuses on whether the network / APP users is a real...

AI Technical Summary

Problems solved by technology

[0007] Generally speaking, for multi-text input, we can splicing the text into a paragraph as a single data point input model learning, but for multiple pictures, simply splicing the pictures will cause a lot of noise, and the performance of the model will be greatly reduced. ;Another method, padding (automatic completion), is to set the number of single-modal models according to the maximum number of pictures, and then automatically fill in the blank pictures for the data with insufficient pictures, which will also cause the decline of model performance and detection time the rise of

Images