Knowledge Graph Weight Adjustment for Relationship Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing knowledge graph systems face challenges in accurately determining relationship strengths between data elements, especially in large graphs, which affects the customizability and accuracy of recommendations.
Innovation Solution
A computer-implemented method that establishes a knowledge graph with nodes and edges representing data elements and their relationships, assigns weights to these elements, and iteratively adjusts modifiable weights to match established relationship strength values, ensuring accurate relationship determination.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If relationship strength values are calculated by summing weights of all paths between nodes, then measurement precision of relationship strength is improved, but device complexity and computational time increase
Solution Approach 1:
The patent applies partial action by calculating relationship strength only for paths up to a maximum length threshold, rather than summing all possible paths of any length. This selective approach maintains measurement precision for significant relationships while avoiding the computational complexity of evaluating excessively long paths that contribute minimally to the relationship strength.
Solution Approach 2:
The patent performs preliminary action by pre-calculating and storing path information between nodes before computing relationship strengths. This includes identifying all paths up to the maximum length threshold in advance, which reduces the computational burden during the actual relationship strength calculation phase.
2Loss of information
If the knowledge graph includes all possible paths between nodes, then completeness of relationship analysis is improved, but loss of time in processing increases
Solution Approach 1:
The patent implements partial action by considering only paths up to a maximum length threshold in the relationship strength calculation. This ensures that the most significant relationships (represented by shorter paths) are fully analyzed while avoiding the time-consuming processing of excessively long paths that have diminishing relevance.
Solution Approach 2:
The patent applies parameter changes by introducing a maximum path length parameter that controls the scope of path evaluation. This parameter can be adjusted to balance between completeness of analysis and processing time, allowing the system to adapt to different computational resource constraints and accuracy requirements.
3Device complexity
If weights are fixed for all graph elements, then device complexity is reduced, but adaptability of the knowledge graph decreases
Solution Approach 1:
The patent applies local quality by allowing different weight adjustment behaviors for different graph elements. Specifically, certain edge types or paths can have fixed weights while others have modifiable weights that can be adjusted based on training data. This localized approach enables adaptability where needed while maintaining simplicity where appropriate.
Solution Approach 2:
The patent implements dynamics by making weights adjustable and trainable rather than completely fixed. The system allows weights to be dynamically modified during training to match established relationship strength values, enabling the knowledge graph to adapt to specific applications and domains while maintaining a structured framework.
4Manufacturing precision
If iterative weight adjustment is performed to match established relationship strength values, then manufacturing precision of relationship determination is improved, but loss of time in computation increases
Solution Approach 1:
The patent applies feedback by using the difference between calculated and established relationship strength values to guide weight adjustments. The system iteratively modifies weights based on this feedback signal, progressively reducing the error and improving accuracy. This feedback-driven approach ensures systematic convergence toward the target relationship strengths.
Solution Approach 2:
The patent performs preliminary action by pre-processing the training data to extract established relationship strength values before the iterative training process. This preparation work is done in advance to reduce the computational burden during iterative weight adjustment, as the target values are readily available for comparison without requiring complex calculations during each training iteration.
Data Source
AI summary
Some embodiments are directed to a computer implemented system and method to determine relationships between data elements using a knowledge graph are described. The method comprises establishing a knowledge graph having graph elements, associating weights with data elements and determining at least some of these weights to be modifiable weights. Training data is used to determine the relationships between data elements in the knowledge graph by calculating the strength of the relationship. The calculated relationship strength values are compared to previously established relationship strength values, and one or more modifiable weights are modified to reduce the error between the calculated and established relationship strength values. These steps are repeated until a matching criterion is reached.