Electronic Device Keyword Abstraction for Private Recommendations

Solution Overview

Problem

Existing content recommendation models face challenges in balancing resource consumption and personal data security, with training on devices consuming resources and exposing personal data during transmission, or requiring extensive communication.

Innovation Solution

An electronic device extracts keywords from context and content information, converts them into upper concepts based on a predetermined keyword level, and transmits these to a server for model training, reducing resource consumption and protecting personal data.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If the recommendation model is trained on the electronic device using personal data, then content recommendation accuracy is improved, but device resources are consumed and personal data security is compromised

Engineering Contradiction:

Improvecontent recommendation accuracyVSAvoiddevice resource consumption

Core Design Contradiction:

Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The patent extracts only the essential features from personal data to create user embeddings, rather than using raw personal data directly for training. This extraction process removes unnecessary information while preserving the key characteristics needed for accurate recommendations, thereby reducing resource consumption and security risks while maintaining recommendation accuracy.

Inventive Principle:

Principle #2Taking out (Extraction)

Solution Approach 2:

The patent introduces user embeddings as an intermediary representation between personal data and the recommendation model. These embeddings serve as a compressed, abstracted form of user information that can be efficiently processed by the model without requiring extensive computational resources or exposing sensitive personal data.

Inventive Principle:

Principle #24Intermediary (Mediator)

2Measurement precision

If the recommendation model is trained on the electronic device using personal data, then content recommendation accuracy is improved, but personal data security is compromised during data transmission

Engineering Contradiction:

Improvecontent recommendation accuracyVSAvoidpersonal data exposure

Core Design Contradiction:

Measurement precisionVSObject-affected harmful factors

Solution Approach 1:

The patent extracts only the essential features from personal data to create user embeddings, rather than using raw personal data directly for training. This extraction process removes unnecessary information while preserving the key characteristics needed for accurate recommendations, thereby reducing resource consumption and security risks while maintaining recommendation accuracy.

Inventive Principle:

Principle #2Taking out (Extraction)

Solution Approach 2:

The patent creates a simplified copy of user information in the form of user embeddings. These embeddings are mathematical representations that capture user preferences and characteristics without containing actual personal data, thus enabling model training without exposing sensitive information.

Inventive Principle:

Principle #26Copying

3Ease of operation

If the recommendation model is loaded on both the electronic device and the server, then content recommendation service is provided, but communication resources are consumed for exchanging training results

Engineering Contradiction:

Improvecontent recommendation serviceVSAvoidcommunication resource consumption

Core Design Contradiction:

Ease of operationVSLoss of energy

Solution Approach 1:

The patent extracts only the essential features from personal data to create user embeddings, rather than using raw personal data directly for training. This extraction process removes unnecessary information while preserving the key characteristics needed for accurate recommendations, thereby reducing resource consumption and security risks while maintaining recommendation accuracy.

Inventive Principle:

Principle #2Taking out (Extraction)

Data Source

PatentUS12418700B2Electronic device for providing content recommendation service, and method therefor

Publication Date: 2025.09.16 SAMSUNG ELECTRONICS CO LTD

AI summary

An electronic device may include a wireless communication circuit, a memory, and a processor. The memory may store instructions that allow, when executed, the processor to: obtain context information associated with the state of the electronic device, and contents information being provided by the electronic device; extracting first keywords on the basis of the context information and the contents information; analyze the keyword level for the first keywords; changing the first keywords into a superordinate concept according to preset keyword levels to obtain second keywords; and transmit the second keywords to at least one server by using the wireless communication circuit. Other various embodiments identified through the specification are possible.