A method for determining the online behavior category of network users

Abstract

The present invention proposes a method for determining the online behavior category of network users, the method comprising: extracting the online behavior characteristics of each network user to be tested, and forming a user behavior characteristic matrix X through the quantization method of the document vector space model; The user behavior feature matrix X, through the probabilistic latent semantic analysis method PLSA and EM algorithm, obtains the behavior tendency set T and the "user-tendency" probability distribution matrix D; according to the user behavior feature matrix X, through the support vector machine SVM algorithm , to get the probability distribution matrix C of "characteristic word-category"; run T×C through matrix multiplication to get the mapping matrix M of "propensity-category"; run D×M through matrix multiplication to get the probability distribution matrix Y of "user-category"; according to any A probability distribution of network users to be tested in each category, and any network user to be tested is classified into the category with the highest probability value.

Description

technical field [0001] The invention relates to the technical field of the Internet, in particular to a method for determining the online behavior category of network users. Background technique [0002] A large number of cases show that the level of network security management can be effectively improved by using user behavior category information. However, based on the behavior data of web users, it is very difficult to accurately predict the class probability distribution of users by relying on existing techniques. On the one hand, the traditional classification method is to label the sample set, but because it is very difficult or even infeasible to label some users, the classifier based on it is often not accurate enough and the prediction effect is not ideal. On the other hand, traditional clustering methods can only divide users into one cluster, and a user may have multiple types of behavior tendencies. Therefore, traditional clustering-based classification methods ...

AI Technical Summary

Problems solved by technology

However, based on the behavior data of network users, it is very difficult to accurately predict the user's category probability distribution by relying on existing techniques.

On the one hand, the traditional classification method is to label the sample set, but because it is very difficult or even infeasible to label some users, the classifier based on it is often not accurate enough and the prediction effect is not ideal

On the other hand, the traditional clustering method can only divide users into one cluster, and a user may have multiple categories of behavior tendencies, so the classification method based on traditional clustering cannot reflect the real category distribution of network users

Images