Federation recommendation gradient acquisition method and device, intelligent terminal and storage medium

Abstract

The invention discloses a federation recommendation gradient obtaining method and device, an intelligent terminal and a storage medium, and the federation recommendation gradient obtaining method comprises the steps: obtaining common objects and denoising objects; based on the model parameters, obtaining a first noise-containing gradient through each common object; based on the model parameters, obtaining a second noise-containing gradient through each denoising object; and based on the first noise-containing gradient and the second noise-containing gradient, eliminating the common object gradient noise and the denoised gradient noise, and obtaining a target gradient. In this way, the first noise-containing gradient and the second noise-containing gradient both contain corresponding gradient noise, and the scoring behavior of the user can be protected; meanwhile, corresponding gradient noise can be eliminated after the first noise-containing gradient and the second noise-containing gradient are obtained, and the target gradient without the gradient noise is obtained. Therefore, according to the scheme, gradient noise can be eliminated while user scoring behaviors are protected, andthe accuracy of the model in the federation recommendation process is improved.

Description

technical field [0001] The present invention relates to the technical field of federated recommendation, in particular to a federated recommendation gradient acquisition method, device, intelligent terminal and storage medium. Background technique [0002] With the improvement of people's awareness of privacy protection, traditional system filtering algorithms and other methods that collect user data to the server and then model them are no longer feasible because they will leak user privacy. Google first proposed federated learning to solve user privacy issues. Some studies further combined federated learning with collaborative filtering algorithms in recommendation systems to ensure that user data is kept locally during the modeling process, and only model parameters are uploaded to the server. To update the model, the federated recommendation. [0003] In the prior art, in federated recommendation, the gradients of virtual items not rated by the user and the gradients of...

AI Technical Summary

Problems solved by technology

[0006] Aiming at the prior art, the gradient of the virtual user’s unrated items and the gradient of the user’s real rated items are uploaded to the server together, and the gradient noise is introduced while protecting the user’s rating behavior, which reduces the accuracy of the model and the recommendation effect. Problem, the present invention provides a federated recommendation gradient acquisition method, device, intelligent terminal and storage medium, which can acquire model parameters; acquire common objects and denoising objects; based on the above model parameters, acquire the first noisy gradient through each of the above common objects ; Based on the above model parameters, obtain the second noise-containing gradient through each of the above-mentioned denoising objects; eliminate the above-mentioned common object gradient noise and the above-mentioned denoising gradient noise based on the above-mentioned first noise-containing gradient and the above-mentioned second noise-containing gradient, and obtain the target gradient

Images