Ontology-Guided Knowledge Graph Embedding Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing knowledge graph embedding models struggle to effectively answer queries over incomplete knowledge graphs, as they lack a systematic approach to minimize the distance between query embeddings and their answers, especially when ontological axioms are involved.
Innovation Solution
A computer-implemented method for training a knowledge graph embedding model enhanced by an ontology, which involves strategic sampling of queries based on ontology-guided query shapes, generalizations, and specializations to minimize the distance between query embeddings and their answers, while maximizing the distance from non-answer embeddings, using an extended loss function that accounts for ontological axioms.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If knowledge graph embedding models are trained with random query sampling, then the training process is simple and fast, but the model cannot systematically minimize the distance between query embeddings and their answers, especially for incomplete knowledge graphs
Solution Approach 1:
The patent applies preliminary action by pre-computing and storing ontology-guided query shapes, generalizations, and specializations before training. The system prepares training queries in advance using the ontology structure, so that during training, the model can systematically access these pre-organized queries to minimize embedding distances without requiring complex real-time query generation
Solution Approach 2:
The patent segments the training process into distinct components: ontology processing, query shape definition, generalization generation, and specialization generation. Each component handles a specific aspect of the training data preparation, allowing the system to manage complexity through modular organization while achieving systematic embedding distance minimization
2Adaptability or versatility
If the training data is expanded to include ontology-based generalizations and specializations, then the model can better handle incomplete knowledge graphs, but the computational effort and data processing time increase
Solution Approach 1:
The system performs preliminary computation of ontology-based generalizations and specializations before the actual training begins. By pre-generating these training queries using the ontology structure, the patent reduces the computational burden during training and allows the model to focus on learning embedding relationships rather than generating training data in real-time
Solution Approach 2:
The patent introduces dynamic control mechanisms including a generalization depth parameter and a specialization depth parameter that allow the system to adjust the complexity of generated queries. These parameters enable the training process to adapt to different knowledge graph completeness levels and computational resource constraints, balancing adaptability with training efficiency
3Measurement precision
If the model minimizes distance between query embeddings and answers, then query answering precision improves, but the model may fail to distinguish between answers and non-answers in incomplete knowledge graphs
Solution Approach 1:
The patent applies local quality by treating answer embeddings and non-answer embeddings differently in the training process. The loss function is designed to specifically minimize distances between query embeddings and answer embeddings while simultaneously maximizing distances between query embeddings and non-answer embeddings. This localized differentiation ensures the model learns distinct embedding relationships for correct and incorrect answers, improving both measurement precision and reliability
Solution Approach 2:
The training process incorporates feedback mechanisms through the loss function that continuously monitors and adjusts embedding distances. The system receives feedback from the ontology structure and query-answer relationships, using this information to refine the embedding space. This feedback loop ensures that the model not only minimizes distances for correct answers but also maintains clear distinctions for non-answers, improving overall query answering reliability
Data Source
AI summary
A device, computer program, computer-implemented method for training a knowledge graph embedding model of a knowledge graph that is enhanced by an ontology. The method comprises training the knowledge graph embedding model with a first training query and its predetermined answer to reduce, in particular minimize, a distance between an embedding of the answer in the knowledge graph embedding model and an embedding of the first training query in knowledge graph embedding model, and to reduce, in particular minimize, a distance between the embedding of the answer and an embedding of a second training query in knowledge graph embedding model, wherein the second training query is determined from the first training query depending on the ontology.