Click rate estimation method and related device and system

Abstract

The invention discloses a click rate estimation method and a related device and system, and the method comprises the steps of sequentially extracting sample commodities from a commodity topological structure diagram constructed based on object behavior data, and executing the following operations for the extracted sample commodities: obtaining at least one-order commodity topological sequence of the sample commodities from the commodity topological structure diagram; determining related feature data of the sample commodity according to the independent feature data of each commodity in the commodity topology sequence; obtaining residual characteristic data of the sample commodity according to the related characteristic data of the sample commodity and the independent characteristic data ofthe sample commodity; and inputting the obtained residual characteristic data of each commodity into a click rate prediction model to obtain a click rate prediction score of each commodity. Accordingto the method, the topological structure of the commodity is utilized to construct the related feature expression of the commodity, and the independent specific expression of the commodity is combined, so that the generalization of the embedded vector layer is improved, the accuracy of estimating the click rate of the commodity by the neural network is improved, and the user experience is improved.

Description

technical field [0001] The present invention relates to the field of network technology, in particular to a click rate estimation method and related devices and systems. Background technique [0002] With the rapid development of the network, more and more users purchase goods through the network. In the process of online commodity transactions, there are scenarios where commodities are displayed to users for users to choose. A strategy for displaying commodities in the prior art is to determine the recommended products to users based on the estimated click-through ratio (CTR) of commodities. Products and display them in sorted order. [0003] At present, the deep learning model used to predict the click-through rate of product recommendation is generally divided into two parts: Sparse Embedding (sparse vector) and Dense Net (neural network). The Sparse Embedding part realizes the learning of user behavior characteristics. The product features of the product show long-tail...

AI Technical Summary

Problems solved by technology

[0006] It can be seen that the existing click-through rate prediction learning models cannot realize the prediction learning of click-through rate very well, and the adequacy, relevance, and generalization of learning cannot be well guaranteed, so it is impossible to accurately predict Estimated click-through rate, unable to accurately recommend products that better meet user needs

Images