Continuous Knowledge Graph Embedding for Weighted Edge Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current Knowledge Graph Embeddings (KGE) models are limited in representing numerical attributes and weighted edges, failing to accurately model numerical literals and preserve weight semantics, which restricts their ability to represent complex relationships and predict edge weights with confidence.
Innovation Solution
A continuous KGE model is introduced that supports numerical attributes as weighted edges, incorporating metadata to generate vector representations of relations, predicting edge weights with confidence scores, and identifying new links between entities, while preserving numerical properties and extending to incorporate multiple modalities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If traditional Knowledge Graph Embeddings (KGE) models are used, then the model structure is simple and easy to implement, but the ability to represent numerical attributes and weighted edges is limited, failing to preserve weight semantics
Solution Approach 1:
The patent transforms the traditional binary edge representation into a continuous weighted edge representation by introducing numerical parameters. The edge weight is modeled as a continuous parameter that can take any real value, allowing the knowledge graph to represent varying degrees of relationships. This parameter change enables the model to capture nuanced semantic information while maintaining a relatively simple overall structure.
Solution Approach 2:
The patent extends the traditional knowledge graph embedding from a discrete, categorical space to a continuous numerical space. By adding the dimension of continuous weight values, the model moves from representing only the presence/absence of relationships to representing the strength and nature of relationships. This dimensional extension allows for more expressive power without significantly complicating the underlying embedding architecture.
2Measurement precision
If continuous KGE model with numerical attributes is implemented, then accurate prediction of weighted relations with confidence scores is enabled, but the computational complexity increases
Solution Approach 1:
The patent replaces traditional statistical or heuristic methods for weight prediction with a neural network-based continuous embedding model. The mechanical substitution involves using differentiable neural network operations to compute edge weights and confidence scores, enabling end-to-end training and optimization. This substitution provides more accurate predictions while the computational overhead is managed through efficient embedding techniques and gradient-based optimization.
3Productivity
If traditional KGE models are used, then the computational resources and time required are low, but the ability to identify new links between entities and handle complex relationships is restricted
Solution Approach 1:
The patent performs preliminary embedding of all entities and relations into continuous vector spaces before actual link prediction tasks. This pre-computation of embeddings allows for rapid inference and prediction of new links, as the heavy lifting of representing complex relationships is done in advance. The continuous embeddings capture intricate relationship patterns that can be quickly queried and extended to predict new connections.
Data Source
AI summary
Embodiments are provided for providing a continuous knowledge graph in a computing system by a processor. One or more weighted values of an edge between a pair of entities in a knowledge graph may be predicted based on one or more candidate statements. A confidence score may be generated for the one or more predicted weighted values.