On-Device Knowledge Graph for Content Recommendation

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing on-device knowledge graph-based recommendation systems struggle to reflect user behavior and context accurately, leading to unsuitable content recommendations and inefficient computational and memory usage.

Innovation Solution

A method and apparatus that generate and update a knowledge graph on-device by analyzing log history information, combining it with a server-generated knowledge graph, and using pattern recognition to provide user-specific content recommendations.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If a knowledge graph is generated and updated on-device based on log history information, then user behavior patterns are reflected accurately in content recommendations, but computational effort and memory usage increase

Engineering Contradiction:

Improveuser behavior pattern recognition accuracyVSAvoidcomputational effort

Core Design Contradiction:

Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The knowledge graph is divided into two segments: a server-generated knowledge graph containing general knowledge and an entity-relationship structure, and a device-generated knowledge graph containing user-specific behavior patterns. This segmentation allows the device to focus computational resources only on generating and updating the user-specific portion, rather than maintaining the entire knowledge graph, thereby reducing computational effort while preserving accurate user behavior pattern recognition.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The server pre-generates the knowledge graph structure including entities, relationships, and attributes before transmission to the device. This preliminary action eliminates the need for the device to perform complex knowledge graph construction from scratch, reducing on-device computational effort while enabling accurate user behavior pattern matching against the pre-established structure.

Inventive Principle:

Principle #10Preliminary action

2Measurement precision

If a knowledge graph is generated and updated on-device based on log history information, then user behavior patterns are reflected accurately in content recommendations, but memory usage increases

Engineering Contradiction:

Improveuser behavior pattern recognition accuracyVSAvoidmemory usage

Core Design Contradiction:

Measurement precisionVSQuantity of substance

Solution Approach 1:

The knowledge graph data is segmented between server and device storage. The server maintains the complete knowledge graph including all entities, relationships, and attributes, while the device stores only the user-specific behavior pattern data. This segmentation significantly reduces the memory footprint on the device while preserving accurate user behavior pattern recognition capabilities.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The user-specific behavior pattern information is extracted from the complete knowledge graph and stored separately on the device. By taking out only the relevant user-specific portion from the full knowledge graph, the system achieves accurate user behavior pattern recognition with minimal memory usage on the device.

Inventive Principle:

Principle #2Taking out (Extraction)

3Adaptability or versatility

If server-generated knowledge graph and device-generated knowledge graph are separately managed, then knowledge graph can be continuously updated on-device, but system complexity increases

Engineering Contradiction:

Improveknowledge graph update capabilityVSAvoidsystem complexity

Core Design Contradiction:

Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The server-generated knowledge graph and device-generated knowledge graph are merged into a unified recommendation system. The device knowledge graph (containing user behavior patterns) is combined with the server knowledge graph (containing general knowledge) to produce personalized content recommendations. This merging approach enables continuous on-device updates while managing system complexity through a clear hierarchical relationship between the two knowledge graphs.

Inventive Principle:

Principle #5Merging (Combining)

Solution Approach 2:

The device-generated knowledge graph acts as an intermediary layer between the server-generated knowledge graph and the recommendation engine. This intermediary structure allows continuous on-device updates of user behavior patterns without requiring changes to the server-generated knowledge graph, simplifying the update mechanism while maintaining adaptability.

Inventive Principle:

Principle #24Intermediary (Mediator)

Data Source

PatentUS11874885B2Method and apparatus for providing content based on knowledge graph

Publication Date: 2024.01.16 SAMSUNG ELECTRONICS CO LTD

AI summary

A content recommendation is provided based on a knowledge graph using, for example, a processor and memory of a device. In some embodiments, sensors are used. The content recommendation efficiently makes user of the processor, memory and/or sensors. A knowledge graph is maintained at the device and another knowledge graph may be maintained at a server. The knowledge graph at the device is generated, for example, by obtaining log history information, generating a device knowledge graph, generating a pattern knowledge graph associated with a behavior pattern of a user, and/or updating the device knowledge graph by adding the pattern knowledge graph to the device knowledge graph. In some examples, the knowledge graph at the server is generated by a server, based on the pattern knowledge graph.