Click rate estimation method, device and equipment and readable storage medium

Abstract

The invention discloses a click rate estimation method, which performs click rate estimation independently according to attribute information of clickable items through at least two different types of estimation models, fuses various models, extracts features through different methods, can effectively improve the representation ability and generalization effect of the models, and improves the accuracy of click rate estimation. After the estimation results output by the estimation models respectively are obtained, the estimation weights corresponding to the estimation results respectively are determined, the importance of the estimation results of different characteristics is distinguished through the weights, the estimation results of the models are subjected to weighted fusion, the estimation contributions of the different models can be fully utilized to carry out proper result fusion, and the estimation accuracy is improved. And the accuracy of the estimation result is improved. The invention further discloses a click rate estimation device and equipment and a readable storage medium, which have corresponding technical effects.

Description

technical field [0001] The present invention relates to the technical field of information processing, in particular to a click rate estimation method, device, equipment and readable storage medium. Background technique [0002] After years of popularization and development of the Internet, clickable items on the Internet such as advertisements, links, pictures, and articles exist in various forms in various Internet products and have become an important part of the Internet. [0003] The core of precise Internet marketing is the click-through rate estimation technology. The higher the click-through rate, the more interested the user is in the clickable item, and the more likely it is to convert into a transaction. Therefore, the prediction of the click-through rate through the model algorithm can be used to guide the accurate delivery of current Internet clickable items such as advertisements, links, pictures, and articles, so as to maximize the economic benefit output. ...

AI Technical Summary

Problems solved by technology

The single deep learning model method often uses the classic DNN (deep neural network) method to predict the click-through rate. This type of model uses deep learning to have a good representation ability for data, but the single linear model has poor generalization ability, resulting in Model accuracy is not high

Hybrid model methods mainly include DeepFM (wide and deep model) method, AFM (Attentional Factorization Machine) model, NFM (Neural Factorization Machine) model, etc. The hybrid model combines the advantages of traditional machine learning methods and deep learning methods, but its integration process The degree of optimization is limited, which affects the effect of the final model estimation

Images