Knowledge Graph Expansion via Dual Neural Network Confidence
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for expanding knowledge graphs from unstructured text data are inefficient in extracting and validating new triplets, leading to potential noise and degradation in graph quality.
Innovation Solution
An electronic device is configured to generate training and validation data from original text, train neural networks to extract and validate triplets, and apply confidence weights to expand the knowledge graph, ensuring precision and recall thresholds are met, and noise is minimized.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If single neural network is used to extract triplets from text data, then extraction speed is improved, but the reliability of extracted triplets deteriorates due to insufficient validation
Solution Approach 1:
The patent divides the triplet extraction and validation process into two separate neural networks: a first neural network for extraction and a second neural network for validation. This segmentation allows each network to specialize in its function, improving both the speed of extraction and the reliability of validation without compromising the other.
Solution Approach 2:
The patent introduces confidence scores as an intermediary mechanism between the first neural network's extraction output and the final knowledge graph expansion. The confidence scores from both networks serve as intermediaries that mediate the decision-making process for adding triplets to the knowledge graph, ensuring reliability while maintaining efficiency.
2Reliability
If multiple neural networks are used to extract and validate triplets, then the reliability of extracted triplets is improved, but the device complexity increases
Solution Approach 1:
The patent designs both neural networks to process the same input format (text data) and output the same structure (triplets), creating a universal pipeline where the first network extracts and the second network validates using consistent representations. This universality reduces complexity compared to using completely different systems for extraction and validation.
Solution Approach 2:
The patent changes the parameters and architecture of the neural networks based on their specific functions - the first network is optimized for extraction speed while the second is optimized for validation accuracy. This parameter optimization allows each network to be simpler and more efficient for its specific task rather than requiring both to be overly complex.
3Measurement precision
If confidence thresholds are set high to ensure quality, then the precision of knowledge graph expansion is improved, but the quantity of added triplets decreases
Solution Approach 1:
The patent applies partial action by using confidence thresholds that allow selective addition of triplets based on their confidence scores. Instead of requiring all triplets to meet a single high threshold, the system accepts varying levels of confidence appropriate to each triplet's quality, thereby maintaining precision while capturing more triplets overall.
Solution Approach 2:
The patent dynamically adjusts confidence thresholds and weighting parameters based on the specific task requirements and data characteristics. This parameter flexibility allows the system to optimize the balance between precision and quantity by adapting thresholds rather than using fixed rigid values.
Data Source
AI summary
A method of expanding a knowledge graph and an electronic device for performing the method are provided. The electronic device includes a processor and the processor is configured to train a first neural network to extract the triplet using the training data, to compare quality of the trained first neural network to a threshold value using the validation data, to extract a new triplet by inputting the text data to the trained first neural network, to measure a first confidence of the new triplet using the trained first neural network, to measure a second confidence of the new triplet using a trained second neural network using a triplet labeled to the training data and a triplet labeled to the validation data, and to expand the knowledge graph based on the first confidence and the second confidence.