Putting method, device and equipment for advertisement recommendation system and storage medium

Abstract

The embodiment of the invention discloses a putting method, device and equipment of an advertisement recommendation system and a storage medium. The putting method for the advertisement recommendationsystem and the advertisement recommendation system are used for pushing advertisements to users, and the advertisement recommendation system comprises a plurality of advertisement recommendation positions. The method comprises the following steps: obtaining first training data used for training the advertisement recommendation system, wherein the first training data comprises a plurality of second training data corresponding to different users; counting second training data corresponding to each advertisement recommendation position; performing weighted statistics on the second training datato obtain evaluation indexes of the advertisement recommendation system; and determining a delivery strategy of the advertisement recommendation system according to the evaluation index. The putting method of the advertisement recommendation system provided by the embodiment of the invention can more accurately measure whether the effect of the model after model optimization is improved or not, reduce the online times of the model, save time, also can avoid missing useful optimization, and enables the model iteration speed to be increased.

Description

technical field [0001] Embodiments of the present invention relate to advertisement recommendation technology, and in particular to a delivery method, device, equipment and storage medium of an advertisement recommendation system. Background technique [0002] When developing an algorithm model to solve a business problem, it is often necessary to have an indicator to measure the effect of the model we developed, or when optimizing the model, an indicator is needed to measure the difference between the optimized model and the original model, which is currently commonly used The model evaluation indicators include AUC, logloss, PRC, recall rate, accuracy rate, F1 value and so on. [0003] In the scenario of advertisement recommendation, what we need to do is to make personalized recommendations to users and improve the click-through rate of advertisements. The commonly used model evaluation indicators measure the global ranking ability of the model, and cannot accurately mea...

AI Technical Summary

Problems solved by technology

Commonly used model evaluation indicators measure the global ranking ability of the model, and cannot accurately measure the personalized ranking ability

When the model is optimized to a certain level, indicators such as AUC and PRC will be distorted. It often happens that some optimization of the model improves the AUC indicator, but the click-through rate of the advertisement does not increase after the launch

The whole process of going online and observing the effect will be relatively long, which greatly affects the speed of model iteration

Images