User tag processing method and device and electronic equipment

Abstract

The embodiment of the invention provides a user tag processing method, which comprises the following steps: acquiring user credit data used as a training sample, dividing the user credit data into positive tag user credit data and to-be-marked user credit data according to whether the user credit data carries a tag, determining auxiliary user credit data in the positive tag user credit data, dividing the user credit data into to-be-marked user credit data categories, training a first label prediction model, predicting auxiliary user credit data, determining a prediction value, setting a model threshold, training a second label prediction model in combination with the model threshold, and performing label prediction on newly acquired user credit data by using the second label prediction model. A part of positive label users are used as auxiliary users to train a label prediction first model, so that prediction values corresponding to the auxiliary users can indicate model thresholds of positive sample users missed by the model to a certain extent, and then a label prediction second model is trained according to the model thresholds, so that the marking accuracy can be improved, and missing users are retrieved.

Description

technical field [0001] The present application relates to the field of computers, and in particular to a user label processing method, device and electronic equipment. Background technique [0002] When we manage customers, we often label customers based on their credit data, so that we can classify customer groups in the future. [0003] When marking users, the model is mostly used for prediction. This model for prediction is usually obtained by training with positive and negative samples. However, in reality, most of the data we come into contact with is actually positive and negative samples. Unlabeled negative samples, this is because it is difficult to obtain negative samples, and the negative data is too diverse and dynamic, so it is difficult to obtain better training results. This makes it so that if you hope to collect negative samples for model training and then mark, there will be a problem with too few samples and low marking accuracy. [0004] Therefore, it is...

AI Technical Summary

Problems solved by technology

[0003] When marking users, most of the models are used to make predictions. The models used for prediction are usually trained with positive and negative samples. However, in reality, most of the data we come into contact with are actually positive and negative samples. Unlabeled negative samples, this is because it is difficult to obtain negative samples, and the negative data is too diverse and dynamic, so it is difficult to obtain better training results

This makes it so that if you hope to collect negative samples for model training and then mark, there will be problems with too few samples and low marking accuracy

Images