Knowledge Graph Negative Sample Generation via Ontology Constraints
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Generating negative triples that accurately represent incorrect facts for training knowledge graph embeddings is a challenging problem, as existing methods lack a systematic approach to distinguish between correct and incorrect triples.
Innovation Solution
A computer-implemented method that determines negative samples by predicting triples inconsistent with the knowledge graph and ontology, replacing entities to create semantically similar inconsistent triples, and using these to enhance training data for knowledge graph embeddings.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If negative triples are generated randomly or by simple replacement, then the training data quantity increases, but the semantic consistency and accuracy of negative samples deteriorates
Solution Approach 1:
The patent applies preliminary action by using the trained knowledge graph embedding model to predict potential negative triples before actual training. The system pre-generates candidate negative samples by replacing entities in positive triples, then uses the embedding model to score and filter these candidates, ensuring high-quality negative samples are selected before the training process begins.
Solution Approach 2:
The patent introduces an intermediary mechanism - the knowledge graph embedding model itself - to generate and evaluate negative samples. Rather than relying on random generation, the embedding model acts as an intelligent intermediary that predicts plausible but incorrect triples, ensuring the negative samples are semantically meaningful and appropriately challenging for training.
2Measurement precision
If complex methods are used to generate accurate negative samples, then the accuracy of negative samples improves, but the computational complexity and processing time increases
Solution Approach 1:
The patent applies partial action by generating only the necessary number of negative samples required for effective training, rather than exhaustively generating all possible negative triples. The system scores candidate negative samples and selects only those above a certain threshold, avoiding unnecessary computational effort on low-quality candidates.
Solution Approach 2:
The patent changes parameters by using the embedding model's score predictions as a filtering criterion. Instead of using fixed rules for negative sample generation, the system dynamically adjusts which triples become negative samples based on their predicted scores, optimizing the balance between accuracy and computational efficiency.
3Productivity
If existing simple replacement methods are used for negative sampling, then the processing speed is fast, but the ability to distinguish correct from incorrect facts deteriorates
Solution Approach 1:
The patent introduces the knowledge graph embedding model as an intermediary that evaluates the quality of negative samples. This intermediary layer quickly scores candidate triples using pre-computed embeddings, enabling fast filtering of high-quality negative samples without requiring complex verification processes.
Solution Approach 2:
The system performs preliminary computation by pre-training the embedding model on positive triples before generating negative samples. This preliminary action enables rapid scoring and filtering of negative candidates during the actual training process, maintaining high processing speed while ensuring reliability.
Data Source
AI summary
A method for determining negative samples for training a knowledge graph embedding of a knowledge graph enhanced by an ontology including at least one constraint for distinguishing a fact of the knowledge graph from a spurious fact. The method comprises determining embedding predicted triples; determining a set of triples that comprises a triple of the knowledge graph and at least one of the predicted triples that are inconsistent with respect to the ontology; determining from the set of triples a replacement entity for the object entity in the at least one triple of the predicted triples; and determining the negative sample to comprise the relation, the subject entity and the replacement entity, or determining from the subset a replacement entity for the subject entity in the at least one triple of the predicted triples and determining the negative sample to comprise the relation, the object entity, and the replacement entity.