Focused Knowledge Graph Generation for Efficient KGE Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current techniques for identifying training data for knowledge graph embedding (KGE) models are resource-intensive and inefficient, often resulting in incomplete models due to insufficient data selection, leading to the need for costly retraining with supplemented data.
Innovation Solution
A method that processes an external knowledge graph using a graph construction strategy to generate a focused knowledge graph, with scores from machine learning and content analysis models, determining whether to extend or filter the strategy based on context, density, and performance scores, and then trains a KGE model with the modified data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If current techniques are used to identify training data for KGE models, then the model can be trained, but the process is resource-intensive and inefficient
Solution Approach 1:
The patent extracts and selects only the most informative content from the knowledge graph to create a focused training dataset. By using graph construction strategies and scoring mechanisms (context score, density score, balance score), the system identifies and extracts high-value training examples, eliminating the need to process the entire knowledge graph and reducing computing resource consumption while maintaining training effectiveness.
Solution Approach 2:
The patent segments the knowledge graph into different components and evaluates them using multiple scoring dimensions (context, density, balance). This segmentation allows the system to process and select training data in a structured manner, focusing computational resources on the most valuable segments rather than uniformly processing the entire graph.
2Reliability
If insufficient data selection is performed, then the training process is faster, but the model becomes incomplete due to lack of representative data
Solution Approach 1:
The patent implements feedback mechanisms through scoring models that evaluate the quality of selected training data. The context score, density score, and balance score provide feedback on whether the selected data adequately represents the knowledge graph, allowing the system to adjust data selection to ensure model completeness while optimizing training time.
Solution Approach 2:
The patent changes parameters related to data selection by adjusting the thresholds and weights of scoring mechanisms. By optimizing these parameters, the system can control the balance between training data quantity and quality, ensuring the model achieves sufficient accuracy without requiring excessive training time or computational resources.
3Quantity of substance
If the entire knowledge graph is processed for training, then the model has sufficient data, but the computing cost and time increase significantly
Solution Approach 1:
The patent extracts only the essential training data from the knowledge graph by applying graph construction strategies and scoring mechanisms. This extraction process identifies the most informative triples and relationships, creating a focused training dataset that maintains representativeness while dramatically reducing the quantity of data that needs processing, thereby improving computing efficiency.
Solution Approach 2:
The patent applies partial action by selecting only the necessary portion of the knowledge graph for training rather than processing the entire graph. The scoring mechanisms enable the system to identify and select a representative sample that provides sufficient training data without the need to process all available data, achieving the optimal balance between data quantity and processing efficiency.
Data Source
AI summary
A device may process an external knowledge graph (KG) based on a graph construction strategy to generate a focus KG. The device may utilize a graph machine learning model to generate a ranked list of candidate links based on the focus KG, and may process the focus KG, with a graph content analysis model, to generate context, density, and balance scores. The device may process the ranked list of candidate links and a gold standard set of links, with a performance analysis model, to generate a performance score for the ranked list of candidate links, and may generate a modified graph construction strategy based on the context score, the density score, the balance score, and the performance score. The device may generate a list of candidate links based on the modified graph construction strategy, and may train a KG embedding model with the modified ranked list of candidate links.