A User Click Prediction Method Based on Gradient Boosting Decision Tree

Abstract

A user click prediction method based on a gradient boosting decision tree, including establishing a training data set based on the user website access behavior data and related marketing results analyzed by the DPI technology of the telecom operator; preprocessing the data in the training data set; Sort according to the ascending order of the batch numbers of historical sending tasks, select the data of the nearest batch number of historical sending tasks as the verification set, and the data of the remaining batch numbers as the training set; provide the user click prediction model that needs to be established And initialize, use the training set to train the user click prediction model based on the gradient lifting decision tree, and use the verification set to adjust the parameters of the user click prediction model to obtain the final user click prediction model; according to the user click prediction model, predict the user group The click probability value of the task to be delivered by each user, and according to the probability value, the task to be delivered is delivered to the user group according to a predetermined ratio.

Description

technical field [0001] The present invention relates to the technical field of artificial intelligence, and more specifically, to a user click prediction method based on a gradient boosting decision tree. Background technique [0002] In the era of big data, people produce and consume a large amount of information in all aspects of daily life, which makes it difficult for users to find the content they are interested in from a large amount of information. It also becomes difficult for information to effectively reach target users. Therefore, the recommendation system was born in this context. [0003] The main task of the recommendation system is to connect users and information, and provide the most accurate information to users who really need it. For merchants, the recommendation system can provide users with personalized services, improve user trust and stickiness, increase revenue, or accurately place corresponding advertisements to increase revenue. [0004] In the ...

AI Technical Summary

Problems solved by technology

Previous recommendation systems mainly used user’s Internet behavior data and contextual information of advertisements for recommendation and recall. However, since the main data of user behavior in the original data was the URL and number of visits to related websites, compared with the recommendation system of Internet companies For , the amount of information is less, and its prediction is more difficult. Therefore, the effect of this scheme is not very ideal

[0005] In addition, in the CTR estimation scenario based on certain data sets, for example, user website access lines that can be parsed by telecom operators’ Deep Packet Inspection (DPI) technology are used as data sets. However, the data in the data set The value density is still low. On the basis of feature engineering, a suitable algorithm model is still needed to improve the prediction effect

Images