A personalized recommendation method based on federated learning privacy protection

Abstract

The present application relates to the technical field of information security, in particular to a personalized recommendation method based on federated learning privacy protection, which is divided into a federated learning privacy protection stage and a personalized recommendation stage; the federated learning privacy protection stage is used for protecting user privacy, and the personalized recommendation stage is used for reasonably personalized recommendation according to user preferences; a regional server transition containing a user information conversion table is introduced between the two stages, the regional server plays a role of connecting the two stages, the Internet platform sends recommendation information to the regional server, the regional server performs user information conversion and then implements recommendation, and the user does not need to directly face the Internet platform; the regional server is mainly used for parameter conversion between the Internet platform and the user. The method can balance personalized recommendation and data privacy protection to some extent, and can provide an effective reference for personalized recommendation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of information security, and in particular to a personalized recommendation method based on federated learning privacy protection. BACKGROUND

[0002] With the development of the Internet, there are more and more ways for users to receive information. How to timely and effectively grasp the part of the user's most interest in the vast amount of information and improve the interactive friendliness of the Internet platform and the user has become a hot topic in Internet research. At present, the personalized recommendation mechanism is mostly used, that is, the user's preferences are analyzed by collecting and processing local information, historical records, and the time spent on videos and webpages, and then the content is pushed according to the different preferences of different users. In order to accurately push, the personalized recommendation mechanism needs to continuously collect user information to obtain the user's preferences.

[0003] The personalized recommendation system helps users find interesting content in the vast amount of information. The "personalized" service is now increasingly accepted by many fields such as business websites and electronic libraries, and has become an important function. The personalized recommendation algorithm can help users quickly and accurately obtain the content of their interest in hundreds of millions of information, make full use of limited time, and greatly improve people's pursuit of fast-paced life. On the other hand, it can also help the Internet platform improve operational efficiency and attract potential users through personalized recommendation services to improve the value of the Internet platform. At present, most application software uses recommendation algorithms, such as Spotify, Netflix, etc. In China, the recommendation system of NetEase Cloud Music and Douyin has also received good comments from users.

[0004] The recommendation system is usually divided into three parts: user modeling, recommended object modeling, and recommendation algorithm. Whether it is user modeling or recommendation algorithm, effective operation and cooperation cannot be achieved without the collection and processing of user information. In order to ensure the accuracy of the recommendation result, the recommendation system needs to collect a large amount of user information. User information often carries sensitive information of the user. Sensitive information is most closely related to the user, and if this part of information is not properly protected, it will cause harm to the user, and even threaten national security.

[0005] Therefore, how to protect the user's sensitive information from being leaked under the personalized recommendation mechanism has become the focus of data security. The present application provides a personalized recommendation method based on federated learning privacy protection, which can realize personalized recommendation service and protect user privacy. SUMMARY

[0006] In view of the above shortcomings in the prior art, the present application provides a personalized recommendation method based on federated learning privacy protection, which has reasonable structure design and is convenient for promotion.

[0007] A personalized recommendation method based on federated learning privacy protection, which is divided into a federated learning privacy protection stage and a personalized recommendation stage;

[0008] The federated learning privacy protection stage is used for protecting user privacy, and the personalized recommendation stage is used for reasonably personalized recommendation according to user preferences; a regional server transition containing a user information conversion table is introduced between the two stages, the regional server plays a role of connecting the two stages, the Internet platform sends recommendation information to the regional server, the regional server performs user information conversion and then implements recommendation, and the user does not need to directly face the Internet platform; the regional server is mainly used for parameter conversion between the Internet platform and the user.

[0009] As preferred, the specific steps are as follows:

[0010] S1, introducing a real user information pool for storing real user information;

[0011] S2, introducing a virtual user information pool for storing virtual user information generated according to certain rules;

[0012] S3, introducing a user information conversion technology for conversion between real user information and virtual user information;

[0013] S4, introducing a user information conversion table for storing the mapping relationship between real user information and virtual user information during user information conversion; the information cache table includes real user information, virtual user information and update time;

[0014] S5, introducing regional hierarchical federated learning, that is, the recommendation system does not implement personalized recommendation for a specific user, but implements personalized recommendation service for a certain region.

[0015] The beneficial effects of the present application are:

[0016] (1) The present application provides a personalized recommendation method based on federated learning privacy protection, which can balance personalized recommendation and data privacy protection to some extent and provide effective reference for personalized recommendation.

[0017] (2) The sensitive information of the user is protected to avoid privacy information leakage and protect the information security of the user and the country.

[0018] (3) It has a positive reference significance for personalized recommendation of Internet data. BRIEF DESCRIPTION OF DRAWINGS

[0019] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed to be used in the embodiments will be briefly introduced as follows. It should be understood that the following drawings only show some of the embodiments of the present application, and therefore should not be regarded as a limitation on the scope. Other related drawings can also be obtained by those of ordinary skill in the art without creative labor on the basis of these drawings.

[0020] Figure 1 The figure is a schematic diagram of the principle of the present application. DETAILED DESCRIPTION

[0021] In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are some of the embodiments of the present application, but not all the embodiments. The components of the embodiments of the present application described and shown in the drawings herein can be arranged and designed in various different configurations.

[0022] A personalized recommendation method based on federated learning privacy protection, which is divided into a federated learning privacy protection stage and a personalized recommendation stage. The federated learning privacy protection stage is used to protect user privacy, and the personalized recommendation stage is used to make reasonable personalized recommendations according to user preferences. The regional server transition containing the user information conversion table is introduced between the two stages. The regional server plays a role of connecting the past and the future. The Internet platform does not need to directly implement the recommendation function for specific users, but sends the recommendation information to the regional server, which implements the recommendation after converting the user information. Users do not need to directly face the Internet platform, which improves the security of users. The regional server is mainly used for parameter conversion between the Internet platform and the user, and this method can protect the privacy of user data.

[0023] The specific steps are as follows:

[0024] S1, introduce a real user information pool for storing real user information;

[0025] S2, introduce a virtual user information pool for storing virtual user information generated according to certain rules;

[0026] S3, introduce a user information conversion technology for conversion between real user information and virtual user information;

[0027] S4, introduce a user information conversion table for storing the mapping relationship between real user information and virtual user information during user information conversion; the information cache table includes real user information, virtual user information, and update time;

[0028] S5, introduce regional layered federated learning, that is, the recommendation system does not implement personalized recommendation for a specific user, but implements personalized recommendation service for a certain region.

[0029] As shown in the accompanying Figure 1

[0030] ① User accesses the Internet platform.

[0031] ② User information conversion, converts the real information of the user into virtual user information.

[0032] ③ The regional server accesses and browses the Internet platform.

[0033] ④ The Internet platform recommends according to the preferences shown by the regional server.

[0034] ⑤ User information conversion, converts the virtual user information into real user information.

[0035] ⑥ Send the access data to the user.

[0036] Among them, the data does not change in the transmission process, and the only change is the user information field in the transmission process.

[0037] The personalized recommendation method based on federated learning privacy protection can balance personalized recommendation and data privacy protection to some extent, and can provide effective reference for personalized recommendation.

[0038] The above only describes the preferred embodiments of the present application patent, and does not limit the present application patent, any modification, equivalent replacement and improvement made within the spirit and principle of the present application patent should be included in the protection scope of the present application patent.​

Claims

1. A personalized recommendation method based on federated learning privacy protection, characterized in that, The method is divided into a federal learning privacy protection stage and a personalized recommendation stage; The federal learning privacy protection stage is used for protecting user privacy, and the personalized recommendation stage is used for reasonably personalized recommendation according to user preferences; A regional server transition containing a user information conversion table is introduced between the two stages, the regional server plays a role of connecting the two stages, the Internet platform sends recommendation information to the regional server, the regional server implements recommendation after executing user information conversion, and users do not directly face the Internet platform; the regional server is mainly used for parameter conversion between the Internet platform and the users; The specific steps are as follows: S1, introducing a real user information pool for storing real user information; S2, introducing a virtual user information pool for storing virtual user information generated according to certain rules; S3, introducing a user information conversion technology for conversion between real user information and virtual user information; S4, introducing a user information conversion table for storing the mapping relationship between real user information and virtual user information during user information conversion; the information cache table includes real user information, virtual user information and update time; S5, introducing regional layered federal learning, the recommendation system does not implement personalized recommendation for a specific user, but implements personalized recommendation service for a region.

Images