Ontology-Based Knowledge Graph Embedding for New Entity Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Knowledge graph embedding models are unable to predict links associated with new unknown entities without retraining the entire knowledge graph, which is time-consuming and resource-intensive.
Innovation Solution
A prediction platform that generates knowledge graph embeddings using ontological knowledge, allowing for the inclusion of new entities by approximating their embeddings based on existing schema-related and non-schema-related entities, without requiring a full retraining of the knowledge graph.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If knowledge graph embedding models are retrained to include new entities, then prediction accuracy for new entities is improved, but computational time and resource consumption increase significantly
Solution Approach 1:
The system pre-computes and stores embedding vectors for ontology schema elements (classes, properties, relationships) in advance. When a new entity is encountered, its embedding is generated by combining pre-computed schema embeddings rather than retraining the entire knowledge graph, thus avoiding time-consuming full retraining while maintaining prediction accuracy
Solution Approach 2:
The knowledge graph is segmented into schema elements (ontology classes, properties, relationships) and instance data. The embedding model is trained only on schema elements in advance, while new instances can be embedded by combining schema embeddings, separating the heavy training computation from lightweight inference
2Reliability
If full knowledge graph retraining is performed to accommodate new entities, then link prediction reliability is improved, but computational resource consumption increases
Solution Approach 1:
The system performs preliminary training on ontology schema elements and stores the resulting embeddings. New entities are integrated by combining their schema embeddings with instance embeddings, avoiding the need for resource-intensive full graph retraining while preserving link prediction reliability
Solution Approach 2:
Instead of retraining the entire knowledge graph model, the system creates embeddings for new entities by copying and combining relevant schema embeddings (class embeddings, property embeddings, relationship embeddings) to generate accurate entity representations without full retraining
3Adaptability or versatility
If knowledge graph embeddings are updated to include new entities, then system adaptability is improved, but processing complexity increases
Solution Approach 1:
The system segments the embedding generation process into schema-based components (class embeddings, property embeddings, relationship embeddings) and instance components. New entities are processed by combining these segmented embeddings according to their ontology structure, enabling adaptability through modular composition rather than complex full-graph updates
Data Source
AI summary
A device receives training data and an ontology for the training data, where the training data includes information associated with a subject of the ontology. The device generates a knowledge graph based on the training data and the ontology, and converts the knowledge graph into knowledge graph embeddings, where the knowledge graph embeddings include points in a k-dimensional metric space. The device receives a new entity that is not present in the knowledge graph embeddings, and generates a new embedding of the new entity. The device adds the new embedding to the knowledge graph embeddings, and utilizes the knowledge graph embeddings, with the new embedding, to perform an action.