Client-Server Content Recommendation with Local Privacy Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional content recommender systems face privacy concerns due to the requirement for personal data collection and storage, which can lead to unauthorized data access and loss of user control over their information, especially when using public databases.
Innovation Solution
A computer-implemented method that separates private and public content, using a client-server architecture with a firewall to restrict access, and employs a fuzzy query mechanism and local filtering to provide user-controlled privacy settings, maintaining private information on the client system and using transferable model parameters for classification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional content recommender systems collect and store personal data in public databases to improve recommendation accuracy, then recommendation quality is improved, but user privacy control deteriorates
Solution Approach 1:
The patent segments the recommender system into two distinct components: a public server that handles anonymous aggregated data and model training, and a private client that stores and processes personal data locally. This segmentation allows the system to maintain recommendation accuracy through centralized model training while preserving user privacy by keeping personal data confined to individual client devices where users retain full control.
Solution Approach 2:
The patent introduces transferable model parameters as an intermediary between the public server and private client. The server trains models using aggregated anonymous data and generates transferable parameters that can be applied locally at the client without exposing personal data. This intermediary mechanism enables accurate recommendations while maintaining a clear boundary that protects user privacy.
2Adaptability or versatility
If traditional systems store personal data on centralized servers to enable comprehensive analysis, then recommendation capability is improved, but security risks increase
Solution Approach 1:
The system segments data storage and processing functions between centralized and decentralized locations. Personal identifiable information and sensitive user data are stored exclusively on private client devices under user control, while the public server stores only anonymous aggregated data and model parameters. This segmentation maintains recommendation capability through centralized model training while dramatically improving data security by eliminating centralized repositories of personal data.
Solution Approach 2:
The patent extracts personal data from the centralized server environment and relocates it to private client storage. By taking out sensitive information from the vulnerable centralized system and placing it in secure local storage under user control, the system maintains its recommendation capability through model parameters while removing the security vulnerability of centralized personal data storage.
3Measurement precision
If comprehensive personal data is collected to train classification models, then model performance is improved, but computational requirements on client side increase
Solution Approach 1:
The patent applies partial action by having the public server perform the computationally intensive tasks of data processing and model training using aggregated anonymous data. The private client receives pre-trained transferable model parameters and only performs lightweight local inference. This division allows the system to achieve high model performance through comprehensive server-side training while keeping client-side computational requirements minimal.
Solution Approach 2:
Transferable model parameters serve as an intermediary that bridges the computational gap between server-side training and client-side deployment. The server performs extensive computations to train models on aggregated data and encapsulates the results in transferable parameters that can be efficiently applied locally. This intermediary eliminates the need for clients to perform heavy computational training while still achieving high model performance.
Data Source
AI summary
Techniques for recommending content to a user include identifying classified public content stored on a server appliance or a repository; identifying private content of a user stored on a client appliance or a repository, the client appliance communicably coupled to the server appliance through a network; receiving, from the user, a request for a recommendation of content; generating a representative query based on the request for the recommendation of content; determining, based on the representative query, a portion of the classified public content stored on a server appliance or the repository; determining, based on the request, a portion of the private content stored on the client appliance or the repository; and preparing, for presentation to the user, the portion of the classified public content based on the representative query and the portion of the private content based on the request for the recommendation of content.