Knowledge Graph Embedding via Two-Layer Fusion for Sparse Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing knowledge graph embedding representation methods are limited by sparse graph structures and external information features, leading to poor completion effects due to reliance on text corpus scale, which hinders the completeness and accuracy of knowledge graph applications.
Innovation Solution
A knowledge graph embedding representation method that uses a two-layer information fusion mechanism to model entity and relationship embeddings, incorporating semantic correlations and weight coefficients to enhance representation capabilities and improve completion accuracy, by obtaining entities and related concepts from a preset knowledge base and training the embedding representation model to minimize a loss function based on known fact triplets.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing knowledge graph embedding representation methods are used, then the knowledge graph can be completed based on entity/relationship embedding representation, but the representation capability is limited by sparse graph structure and external information features are easily affected by text corpus scale, resulting in poor completion effect
Solution Approach 1:
The patent segments the embedding representation into multiple layers: entity embedding layer, relationship embedding layer, and composite embedding layer. Each layer processes specific aspects of knowledge graph data independently, allowing the system to handle sparse graph structures more effectively by breaking down the complex representation task into manageable segments that can be optimized separately.
Solution Approach 2:
The patent introduces multi-dimensional embedding vectors to represent entities and relationships, transforming the traditional single-dimension representation into a higher-dimensional space. This dimensional expansion allows the model to capture more nuanced semantic relationships and improve completion accuracy by representing complex interactions that cannot be captured in lower dimensions.
2Quantity of substance
If external information features are used for knowledge graph completion, then more information can be incorporated, but the features are easily affected by the scale of text corpus, leading to suboptimal completion results
Solution Approach 1:
The patent introduces attention mechanisms as intermediaries between external information features and the embedding representation. The attention mechanism selectively weights and filters external information based on its relevance to the knowledge graph completion task, preventing noise from large-scale text corpora from degrading the quality of the embedding representation while still incorporating valuable external information.
Solution Approach 2:
The patent dynamically adjusts the weighting parameters of external information features based on the specific completion task and available data. By changing the parameters that control how external features are incorporated, the system can optimize completion accuracy for different scenarios rather than being uniformly affected by text corpus scale.
3Adaptability or versatility
If a two-layer information fusion mechanism is used to model entity/relationship embedding representation, then semantic extension of entities can be implemented and representation capability can be improved, but the model complexity increases
Solution Approach 1:
The patent implements a nested embedding structure where relationship embeddings are nested within entity embeddings, and composite embeddings are nested within the overall knowledge graph representation. This nested doll approach allows the model to capture hierarchical relationships and semantic extensions at multiple levels while maintaining a structured organization that manages complexity through systematic nesting rather than scattered complexity.
Data Source
AI summary
A knowledge graph embedding representation method and a related device are disclosed. The method includes: obtaining, from a preset knowledge base, N related entities of each entity in M entities of a target knowledge graph and K concepts corresponding to each of the N related entities, determining a semantic correlation between each entity and each of the N related entities of the entity, determining a first entity embedding representation of each of the N related entities based on the corresponding K concepts, modeling, based on the first entity embedding representation and the semantic correlation, an entity/relationship embedding representation, and training a model according to an attention mechanism and a preset model training method, to obtain the entity/relationship embedding representation.