Internet authenticity consumer discrimination method, system and device

Abstract

According to the Internet authenticity consumer discrimination method, system and device provided by the invention, a multi-mode abnormal account identification method is designed for social and vertical platforms, the speaking content and behavior of a user are combined, and the method comprises the following steps: randomly selecting a sample user, extracting behavior data of the user and publishing content data; establishing a dynamic threshold value of feature anomaly judgment to determine a quantitative difference between an abnormal behavior and a normal behavior of the user; and establishing an evaluation mechanism according to the dynamic threshold to dynamically evaluate the state of the user, and identifying an abnormal user account.

Description

technical field [0001] The invention relates to the technical field of flow analysis, in particular to a method, system and device for discriminating authentic Internet consumers. Background technique [0002] At present, there are mainly three methods for identifying fake consumers (abnormal network accounts): analysis based on content characteristics, analysis based on user behavior characteristics, and analysis methods based on user relationship characteristics. [0003] Content feature analysis refers to the use of natural language processing related technologies, such as text analysis and sentiment analysis to identify similarities and emotional tendencies between texts, so as to achieve the purpose of identifying abnormal accounts. This technology was widely used mainly in the early Internet era, such as identifying spam by detecting duplicate content. However, with the advancement of the Internet, abnormal accounts will hide themselves more and more, and can even imi...

AI Technical Summary

Problems solved by technology

However, the application of this method based on user characteristics is almost impossible to be realized in commercial applications as the amount of data on social platforms increases (billions, tens of billions).

The reason is that 1) the model needs to manually judge and label which are abnormal accounts as training data. If you want to find out which are abnormal accounts among 100 million pieces of user data, you need to manually label a large amount of data, which is costly

2) Even with marked abnormal account data, the prediction needs to spend a lot of computing resources and time

3) The accuracy of this type of model is usually around 80%, which is still a long way from commercial applications

[0005] Similarly, methods based on user relationship recognition rely on models such as graph theory and probability graphs, and will encounter problems similar to machine learning models in applications.

In addition, as Internet users pay more attention to the protection of personal privacy, the user social network data necessary for the method based on user relationship identification becomes very difficult to obtain, which further increases the difficulty of implementing this method.

Images