CTR model adaptive incremental training method, apparatus and device, and storage medium

Abstract

The invention discloses a CTR model adaptive incremental training method and device, equipment and a storage medium. The method comprises the steps of collecting an initial data set; based on the initial data set, training the CTR model to obtain an initialized model; collecting an incremental data set at preset time intervals, wherein the incremental data set comprises newly added data and historical data; adjusting corresponding parameters in the initialization model according to the incremental data set to obtain an adjusted initialization model; and based on the adjusted incremental data set and the adjusted parameters corresponding to the initialization model, performing incremental training to obtain an incremental model, and taking the incremental model as a new initialization model, thereby solving the problems that in the prior art, the timeliness and the data utilization rate of model training are relatively low and the training efficiency is relatively high due to the fact that total training is performed by selecting data with a fixed time span. And the user experience feeling is poor.

Description

technical field [0001] The present application relates to the technical field of machine learning, and in particular to a CTR model adaptive incremental training method, device, device and storage medium. Background technique [0002] In the context of the rapid development of global e-commerce, personalized search and recommendation systems have become the main means for e-commerce platforms to solve the problem of "information overload" and meet the diverse needs of users. On the one hand, good recommendation results are conducive to satisfying user interests, and on the other hand, they are conducive to the development of the platform. With the development of artificial intelligence technology, the search recommendation system of e-commerce platforms has become relatively mature. At present, most search recommendation systems in the industry include the following two stages: recall and sorting. The recall stage is to select a collection of commodities that meet the user'...

AI Technical Summary

Problems solved by technology

[0004] The present application provides a CTR model self-adaptive incremental training method, device, equipment and storage medium, which are used to solve the problems of timeliness and data utilization of model training caused by selecting data with a fixed time span for full training in the prior art. Relatively low, and technical problems with poor user experience

Images