Method and system for recognizing hidden paid posters on basis of fusion of behavior characteristic and content characteristic

Abstract

The invention relates to a method and a system for recognizing hidden paid posters on the basis of fusion of behavior characteristic and content characteristic. The method includes acquiring initial data including the behavior characteristic and the content characteristic of a user, performing dimensional fusion of the behavior characteristic and the content characteristic by the aid of a Markov Chain Monte Carlo stochastic model so as to form a user characteristic vector, performing DBN(deep belief network) model training according to the user characteristic vector so as to obtain a DBN model, detecting the DBN model to judge whether the detection result meets preset standards or not, and ending if so, otherwise, generating a corresponding regulation order according to the detection result to regulate related parameters at the characteristic fusion stage and the DBN model training stage respectively, and continuously optimizing proportional allocation of the behavior characteristic and the content characteristic, selection of specific characteristics and regulation of iterations during the DBN model training process according to recognition precision rate, thereby achieving the optimal training effect and finally improving recognition precision rate and adaptability of recognition methods.

Description

technical field [0001] The present invention relates to the field of network security, in particular to a water army identification method and system based on the fusion of behavioral features and content features. Background technique [0002] Cyberspace (CyberSpace) has become the fifth-dimensional space for human activities after the four-dimensional space of land, sea, air, and sky. The security of cyberspace will directly affect national security and social development. In recent years, the network security situation has become increasingly severe, and a large part of the threats come from the "Internet Water Army" (HiddenPaidPosters / InternetWaterArmy) that affects the social order of the network anytime and anywhere. -"Internet trolls" generally refer to full-time or part-time groups that post tendentious comments on the Internet for profit. They kidnap public opinion and interfere with public opinion to achieve certain commercial or other purposes. The "Feast" film r...

AI Technical Summary

Problems solved by technology

[0004] However, in the DBN training process, there are mainly two problems: first, the characteristics of the navy are mainly divided into behavioral characteristics and content characteristics, and there is no unified criterion for the selection of each type of characteristics

But how to integrate, specifically select those behavioral features and select those content features to add to the training model, there is currently a lack of effective selection criteria and fusion methods

Second, the dimension of the input features needs to be determined before DBN model training, and the specific selected features cannot be changed during the training process.

However, as the water army changes, the features also change. Therefore, artificially specifying the feature dimension for learning cannot effectively describe feature changes and realize dynamic adaptation of recognition methods.

Images