Graph Neural Network Embeddings for Knowledge Graph Querying

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing question answering over knowledge graph (KG-QA) approaches are inefficient due to the need to search all possible subgraphs and face ambiguity from the structural complexity of natural language questions, as well as a semantic gap between question and KG structures.

Innovation Solution

A graph-based matching and ranking model that encodes graph structure information of knowledge graph subgraphs and question graphs into neural network embeddings using bidirectional graph neural networks, facilitating a match-then-rank approach to identify relevant KG subgraphs for answering natural language questions.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If traditional semantic parsing or retrieve-then-extract methods are used to query knowledge graphs, then the system can process natural language questions, but the search efficiency deteriorates due to the need to search all possible subgraphs

Engineering Contradiction:

ImproveKG-QA processing efficiencyVSAvoidsubgraph search time

Core Design Contradiction:

ProductivityVSLoss of time

Solution Approach 1:

The patent applies preliminary action by pre-computing and storing graph structure information (such as node embeddings, edge features, and subgraph signatures) in the knowledge graph before querying. When a natural language question arrives, the system can quickly retrieve and compare pre-computed structures rather than searching all possible subgraphs from scratch, significantly reducing search time while maintaining processing capability

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent replaces the mechanical exhaustive search approach with a neural network-based matching system. Instead of systematically enumerating and comparing all possible subgraphs, the system uses trained neural networks to directly match question representations with knowledge graph structures, substituting computational mechanics with learned patterns that achieve the same goal more efficiently

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

2Adaptability or versatility

If traditional methods process complex natural language questions, then the system can handle diverse queries, but ambiguity increases due to structural complexity

Engineering Contradiction:

Improvequestion processing capabilityVSAvoidsubgraph matching accuracy

Core Design Contradiction:

Adaptability or versatilityVSReliability

Solution Approach 1:

The patent introduces an intermediary representation layer between natural language questions and knowledge graph structures. The system uses semantic parsers and neural network embeddings as intermediaries that translate questions into graph-compatible representations, bridging the semantic gap and reducing ambiguity while maintaining the ability to handle diverse query types

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The patent applies parameter changes by transforming the representation parameters of natural language questions from raw text to structured graph embeddings. By changing the parameter space from unstructured text to structured vector representations that capture semantic relationships, the system maintains versatility in handling different question types while improving matching reliability through consistent parameter transformations

Inventive Principle:

Principle #35Parameter changes

3Productivity

If traditional approaches query knowledge graphs without incorporating graph structure information, then the system can perform basic queries, but precision deteriorates due to the semantic gap between natural language and KG structures

Engineering Contradiction:

Improvequery execution speedVSAvoidsubgraph identification accuracy

Core Design Contradiction:

ProductivityVSMeasurement precision

Solution Approach 1:

The patent merges graph structure information with natural language processing by combining question embeddings with knowledge graph structural features in a unified neural network model. This merging allows the system to maintain fast query execution while simultaneously improving precision, as the combined representation captures both linguistic meaning and structural relationships

Inventive Principle:

Principle #5Merging (Combining)

Solution Approach 2:

The patent creates a composite representation by combining multiple types of information (natural language semantics, graph structure, node attributes, edge relationships) into a unified embedding space. This composite approach is analogous to composite materials where combining different properties creates a substance with superior characteristics, achieving both speed and precision in KG-QA tasks

Inventive Principle:

Principle #40Composite materials

Data Source

PatentUS12242975B2Querying knowledge graphs with sub-graph matching networks

Publication Date: 2025.03.04 INTERNATIONAL BUSINESS MACHINE CORPORATION

AI summary

Techniques regarding identifying candidate knowledge graph subgraphs in a question answering over knowledge graph task are provided. For example, one or more embodiments described herein can comprise a system, which can comprise a memory that can store computer executable components. The system can also comprise a processor, operably coupled to the memory, and that can execute the computer executable components stored in the memory. The computer executable components can comprise a question answering over knowledge graph component that encodes graph structure information of a knowledge graph subgraph and a question graph into neural network embeddings.