Knowledge Graph Embedding Approximation for Unseen Entity Link Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional knowledge graph embedding models struggle to predict the plausibility of links involving unseen entities, leading to inefficiencies in applications like predicting treatment outcomes for unknown patients without requiring extensive retraining.
Innovation Solution
A method and system for generating approximated embedding vectors of out-of-knowledge-graph entities by constructing descriptory triples and obtaining an embedding matrix, using a set of related elements from the knowledge graph to approximate embeddings for unknown entities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If traditional knowledge graph embedding models are used, then the model structure remains simple, but the model cannot predict link plausibility for unseen entities
Solution Approach 1:
The patent introduces an intermediary component called an 'embedding approximator' that acts as a mediator between the target unseen entity and the knowledge graph embedding space. This approximator generates approximate embedding vectors for unseen entities by leveraging relationships with seen entities, enabling link prediction without requiring the unseen entities to be directly embedded in the original embedding space.
Solution Approach 2:
The patent creates a copy or approximation of the embedding representation for unseen entities. Instead of requiring exact embeddings for all entities, the system generates approximate embedding vectors that replicate the essential semantic information needed for link prediction, allowing the model to handle unseen entities through these copied representations.
2Productivity
If the knowledge graph is extended to include unseen entities, then link prediction capability improves, but retraining time increases
Solution Approach 1:
The patent performs preliminary actions by pre-computing and storing embedding vectors for seen entities in the knowledge graph before encountering unseen entities. When unseen entities need to be processed, the system leverages these pre-computed embeddings and the embedding approximator to quickly generate approximate representations, avoiding the need for extensive retraining when new entities are introduced.
Solution Approach 2:
The patent changes the parameter representation approach by shifting from requiring exact embedding vectors for all entities to using approximate embedding vectors generated through the embedding approximator. This parameter change allows the system to handle unseen entities efficiently without retraining, as the approximator can generate suitable representations on-demand based on relational patterns.
3Measurement precision
If approximate embeddings are generated for unseen entities, then prediction accuracy improves, but computational complexity increases
Solution Approach 1:
The patent segments the entity handling process into two distinct parts: seen entities that have exact embedding vectors stored in the knowledge graph, and unseen entities that receive approximate embedding vectors generated by the embedding approximator. This segmentation allows the system to maintain high prediction accuracy for seen entities while efficiently handling unseen entities through approximation, without uniformly increasing computational complexity across all entities.
Data Source
AI summary
The present disclosure describes methods and systems for generating an approximated embedding of an out-of-knowledge-graph entity based on a knowledge graph. The method includes: receiving a target entity, a dataset associated with the target entity, and an embeddings space of a knowledge graph comprising a set of structured data, wherein the target entity is out of the knowledge graph and the embeddings space includes a set of vectors representing the set of structured data in the embeddings space; selecting a set of elements from the knowledge graph, each element being related to the target entity according to the dataset associated with the target entity; constructing a set of descriptory triples based on the target entity and the set of elements; obtaining an embedding matrix based on the descriptory triples and the embeddings space; and generating an approximated embedding for the target entity based on the embedding matrix.