A method and device for user portray

Abstract

the invention discloses a method for user portray In the case of no label in the original training data, a user is classified by using the high heuristic of application name and the feature selectionalgorithm, and then the label information of the training data set is obtained by iterating a classification result repeatedly so that labeled training data set is constructed, and the more accurateuser portrait is finally realized. The invention discloses a user portary device.

Description

technical field [0001] The invention belongs to the field of big data processing, and in particular relates to a user portrait method and device. Background technique [0002] User portrait refers to the description of user characteristics and attributes through various dimensions, and the analysis and statistical mining of potential value information on these characteristics. User portraits are the data foundation of most Internet companies today. [0003] For the portrait problem, there are two most common methods in the prior art, one is a supervised classification algorithm, and the other is an unsupervised clustering algorithm. [0004] The so-called "supervised" means that in the training data set of the model, the samples are labeled. In such existing technologies, the sample labels in the training data set usually come from known data, such as registration information filled in by users, data exchange, data crawlers, and the like. In many scenarios, it is difficul...

AI Technical Summary

Problems solved by technology

In many scenarios, it is difficult to obtain data with known labels, so if there is no such data with known labels, the model will be helpless and unable to complete the training

For example, in the field of security, some user groups are limited and relatively hidden, and it is generally impossible to obtain a labeled training set. Using a "supervised" algorithm cannot complete the training, which is not conducive to the security deployment of government agencies and security companies.

[0005] Therefore, in the existing technology, there is another way of thinking, which is the so-called "unsupervised" algorithm. Among these algorithms, the clustering algorithm is the most classic. The training data set of this algorithm does not need labels, and this type of algorithm does not require the data set to be labeled. , but there are two problems with the use of unsupervised clustering algorithms: first, the amount of calculation of the clustering algorithm in this scenario is huge, and it is almost impossible to complete, because the number of applications currently on the market is tens of millions, and users The number is also hundreds of millions, and clustering algorithms usually need to load all these data at the same time before they can be calculated, which brings high computational overhead

Second, the clustering algorithm may not be able to achieve the desired effect

[0006] It can be seen that in the absence of data labels, none of the existing technologies can achieve the purpose of good user portraits

Images