current position:Home>SIGIR 2022 | Huawei reloop: self correcting training recommendation system

SIGIR 2022 | Huawei reloop: self correcting training recommendation system

2022-05-15 07:32:51Zhiyuan community

Paper title :

ReLoop: A Self-Correction Continual Learning Loop for Recommender Systems

Thesis link :

https://arxiv.53yu.com/pdf/2204.11165.pdf

The current model training process only obtains the user's feedback as a label , Without considering the errors in the previous recommendation . This paper constructs a self correcting learning cycle for the recommendation system ( be called ReLoop), So as to learn from the previous recommendation mistakes . Build custom loss to encourage each new model version to reduce the prediction error of the previous model version during training .

The core : The prediction results of the previous training are used to restrict the performance of the current round of training from being worse than the previous one , Simple and effective .

copyright notice
author[Zhiyuan community],Please bring the original link to reprint, thank you.
https://en.chowdera.com/2022/131/202205102137569692.html

Random recommended