Personalized Search Requests Using Anonymous Profile Embeddings
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing search methodologies compromise user privacy and efficiency by requiring the sharing of sensitive personal data or manual input to provide personalized content, and anonymization methods may not ensure adequate security or accuracy.
Innovation Solution
A content provision system utilizing user data embedding and content determination machine-learning models to generate anonymized and compressed user profiles, enabling secure and efficient content retrieval without sharing sensitive data, by using user data embedding (UDE) and content determination (CD) machine-learning models to process content retrieval requests.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If sensitive personal data is shared with external servers to provide personalized content, then personalization accuracy is improved, but user privacy and security are compromised
Solution Approach 1:
The patent introduces an intermediary mechanism (federated learning system with local embedding models) that enables personalized content delivery without direct data sharing. The user data embedding model runs locally on the client device, transforming sensitive personal data into local representations that never leave the device, while still enabling the content retrieval platform to provide personalized results through the anonymous identification system.
2Object-affected harmful factors
If traditional anonymization methods are used to protect user data, then privacy concerns are partially addressed, but personalization accuracy is reduced
Solution Approach 1:
The patent extracts only the essential identifying features needed for personalization while leaving sensitive personal data behind on the client device. The user data embedding model transforms raw personal data into compressed local embeddings that capture personalization-relevant patterns without containing identifiable personal information, effectively separating the useful signal from the sensitive data.
3Ease of operation
If manual user input or multiple search iterations are used for personalization, then user control is improved, but time consumption and efficiency worsen
Solution Approach 1:
The patent performs preliminary action by pre-processing user data through the user data embedding model to create local embeddings that capture user preferences and characteristics in advance. This preparation work is done once when the user profile is established, enabling rapid personalized content retrieval through subsequent anonymous identification matching without requiring manual user input or multiple iterative searches.
Data Source
AI summary
Method, systems and computer programs for content provision are provided. A requestor node generates an anonymized and compressed representation of the user profile using a user data embedding machine-learning model inputting user data of a user profile. The content retrieval platform receives the anonymized and compressed representation of the user profile and a content retrieval request and inputs the anonymized and compressed representation of the user profile and the content retrieval request to a content determination machine-learning model to determine content in response to the content retrieval request.