Ontology-Guided Negative Sample Generation for Knowledge Graph Embedding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Generating high-quality negative samples for training knowledge graph embedding models is challenging due to the open-world assumption in knowledge graphs, where missing triples are treated as unknown rather than false, making it difficult to distinguish between correct and incorrect triples.
Innovation Solution
A method and device for automatically generating negative samples by using a knowledge graph embedding model to predict triples, determining vector representations of entities and relations, and applying an ontology with constraints to identify triples that violate these constraints, thereby selecting high-quality negative samples that improve the model's accuracy in predicting missing relations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If negative samples are generated randomly or without domain knowledge, then the training process can proceed, but the accuracy and consistency of the knowledge graph embedding model deteriorates
Solution Approach 1:
The patent introduces an ontology as an intermediary component that mediates between the knowledge graph and the negative sample generation process. The ontology provides domain knowledge constraints that guide the generation of negative samples, ensuring they are both diverse and consistent with domain knowledge. This intermediary structure resolves the contradiction by systematically generating reliable negative samples without requiring complex manual curation.
Solution Approach 2:
The patent performs preliminary action by pre-defining ontology constraints and using them to guide negative sample generation before training begins. The system pre-processes the knowledge graph with ontological constraints to identify valid negative samples in advance, rather than generating them randomly during training. This preliminary structuring improves model accuracy while maintaining a manageable process complexity.
2Quantity of substance
If all possible triples are used as negative samples, then coverage is improved, but the quality and domain consistency deteriorates
Solution Approach 1:
The patent applies local quality by making different parts of the negative sample generation process have different properties. Specifically, the generation process is localized to respect ontological constraints for each triple type, ensuring that negative samples are generated with domain-specific quality requirements applied locally rather than uniformly. This allows diverse negative samples to be generated while maintaining domain consistency through constraint-based filtering.
3Reliability
If manual curation of negative samples is performed, then quality is improved, but time consumption and productivity deteriorates
Solution Approach 1:
The patent implements self-service by enabling the system to automatically generate high-quality negative samples using ontological constraints without requiring manual curation. The ontology-based framework allows the system to self-regulate the quality of negative samples through automated constraint checking and validation, eliminating the need for time-consuming manual processes while maintaining sample quality.
Solution Approach 2:
The patent changes the parameters of negative sample generation by transitioning from random or uniform sampling to constraint-based sampling guided by ontological parameters. This parameter change allows the system to automatically adjust the quality and diversity of negative samples based on domain knowledge encoded in the ontology, achieving high quality samples with improved training efficiency.
Data Source
AI summary
A device, computer implemented method, computer program and non-transitory computer-readable storage, for automatically generating negative samples for training a knowledge graph embedding model, The method includes providing at least one first triple, the first triple is a true triple of a knowledge graph, providing at least one second triple, training the knowledge graph embedding model to predict triples of the knowledge graph depending on a set of triples comprising the at least one first triple and the at least one second triple, determining vector representations of entities and relations with the knowledge graph embedding model, determining a plurality of triples with the vector representations of entities and relations, providing an ontology comprising constraints that characterize correct triples, determining with the ontology at least one triple that violates at least one constraint of the constraints or that violates a combination of at least some of the constraints.