Knowledge Graph Embedding Question Answering

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Question answering over knowledge graphs faces challenges such as varying natural language expressions for predicates, entity name ambiguity, and incomplete domains, making it difficult for machines to accurately identify head entities and predicates in knowledge graphs.

Innovation Solution

The Knowledge Embedding based Question Answering (KEQA) framework jointly recovers head entity, predicate, and tail entity representations in knowledge graph embedding spaces using predicate and head entity learning models, along with a Head Entity Detection model, and employs a joint distance metric that considers relation information to accurately answer simple questions.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If knowledge graph embedding methods represent predicates and entities as low-dimensional vectors to preserve relation information, then the ability to capture semantic meaning is improved, but the difficulty increases because predicates can be expressed in different ways in natural language questions and entity names are ambiguous

Engineering Contradiction:

Improvesemantic meaning captureVSAvoidpredicate and entity identification

Core Design Contradiction:

ReliabilityVSDifficulty of detecting and measuring

Solution Approach 1:

The patent introduces a joint embedding space as an intermediary that simultaneously represents questions, head entities, predicates, and tail entities. This unified embedding space serves as a mediator between natural language questions and knowledge graph structures, enabling the system to bridge the gap between varying natural language expressions and fixed KG predicates through semantic similarity measurement in the shared embedding space.

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The patent creates a universal embedding space that serves multiple functions: representing knowledge graph entities, predicates, and relationships simultaneously, while also accommodating natural language question representations. This multi-functional embedding space can handle various question types and expressions, making the system adaptable to different query formulations without requiring separate processing mechanisms for each case.

Inventive Principle:

Principle #6Universality (Multi-functionality)

2Adaptability or versatility

If the system uses entity name ambiguity and partial names to expand search coverage, then the adaptability to different questions is improved, but the number of possible answers increases making accurate identification difficult

Engineering Contradiction:

Improvequestion handling capabilityVSAvoidhead entity and predicate identification accuracy

Core Design Contradiction:

Adaptability or versatilityVSMeasurement precision

Solution Approach 1:

The patent replaces traditional mechanical string-matching approaches with semantic similarity measurement in embedding space. Instead of relying on exact name matches or simple substring matching, the system uses cosine similarity or other distance metrics in the joint embedding space to measure semantic proximity between question representations and knowledge graph triples, enabling accurate matching even when names are ambiguous or partially specified.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent transforms the matching process from discrete string comparison to continuous semantic similarity measurement. By representing entities and relations as continuous vector embeddings and measuring similarity through dot products or cosine similarities, the system can handle partial matches and ambiguous names gracefully, ranking candidates by semantic proximity rather than requiring exact matches.

Inventive Principle:

Principle #35Parameter changes

3Productivity

If the system answers simple questions using knowledge graph embeddings, then the productivity of question answering is improved, but the complexity of the system architecture increases

Engineering Contradiction:

Improvequestion answering efficiencyVSAvoidmodel architecture complexity

Core Design Contradiction:

ProductivityVSDevice complexity

Solution Approach 1:

The patent merges multiple components into a unified architecture: the question encoding module, head entity prediction module, predicate prediction module, and tail entity prediction module all operate within a single joint embedding space framework. This integration allows the system to handle multiple tasks (question understanding, entity linking, relation identification) simultaneously using shared computational resources and unified parameters, reducing overall system complexity while maintaining high productivity.

Inventive Principle:

Principle #5Merging (Combining)

Data Source

PatentUS11727243B2Knowledge-graph-embedding-based question answering

Publication Date: 2023.08.15 BAIDU USA LLC

AI summary

Described herein are embodiments for question answering over knowledge graph using a Knowledge Embedding based Question Answering (KEQA) framework. Instead of inferring an input questions' head entity and predicate directly, KEQA embodiments target jointly recovering the question's head entity, predicate, and tail entity representations in the KG embedding spaces. In embodiments, a joint distance metric incorporating various loss terms is used to measure distances of a predicated fact to all candidate facts. In embodiments, the fact with the minimum distance is returned as the answer. Embodiments of a joint training strategy are also disclosed for better performance. Performance evaluation on various datasets demonstrates the effectiveness of the disclosed systems and methods using the KEQA framework.