Iterative Knowledge Graph Link-Set Prediction for Cross-Domain Solutions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing knowledge graphs are often siloed and fail to effectively address cross-domain challenges in industrial operations, making it difficult to identify and track intricate cross-domain relationships that contribute to issues like low production efficiency, excessive development cycles, and lack of adherence to specifications.
Innovation Solution
A cross-domain intelligent system using a knowledge graph and graph machine learning approach to predict and explain sets of links between operational challenges and solutions, incorporating user feedback to optimize and expand its knowledge base.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If knowledge graphs are maintained as separate siloed domains, then domain-specific knowledge accuracy is preserved, but cross-domain relationship identification capability deteriorates
Solution Approach 1:
The patent merges multiple siloed knowledge graphs into a unified cross-domain knowledge graph that integrates domain-specific knowledge while enabling cross-domain relationship identification. The system combines knowledge from different domains (e.g., engineering, supply chain, manufacturing) into a single graph structure that preserves local domain accuracy while revealing global cross-domain patterns through unified graph machine learning processing.
Solution Approach 2:
The unified knowledge graph serves multiple functions simultaneously: it maintains domain-specific knowledge accuracy for individual domains while also enabling cross-domain relationship analysis, pattern recognition, and holistic operational optimization. The graph structure is designed to be universally applicable across different industrial domains while adapting to domain-specific requirements.
2Measurement precision
If link prediction is performed at single problem-solution level, then prediction precision is improved, but ability to address complex cross-domain challenges deteriorates
Solution Approach 1:
The patent segments the link prediction process into two levels: first performing precise single-link predictions between individual problems and solutions, then aggregating these predictions to identify sets of links that collectively address cross-domain challenges. This hierarchical segmentation allows the system to maintain high precision at the individual link level while building comprehensive cross-domain solution sets.
Solution Approach 2:
The system transitions from single-dimensional single-link prediction to multi-dimensional link set prediction by considering not only individual problem-solution pairs but also the relationships between multiple links, their contextual dependencies, and cross-domain patterns. This dimensional expansion enables the system to address complex cross-domain challenges while maintaining prediction precision through the underlying single-link prediction accuracy.
3Device complexity
If user feedback is not systematically incorporated, then system simplicity is maintained, but knowledge base optimization and expansion deteriorates
Solution Approach 1:
The patent implements a systematic feedback mechanism where user interactions (clicks, views, corrections, ratings) on predicted links are automatically captured and fed back into the graph machine learning model. This feedback loop continuously optimizes the knowledge base by retraining the model with real-world user data, improving prediction accuracy and expanding the knowledge base with validated cross-domain relationships while maintaining a relatively simple system architecture through automated processing.
Data Source
AI summary
This application relates generally to intelligent and explainable link prediction in knowledge graph systems that automatically incorporate user feedback. In one aspect, this application discloses an iterative process for predicting a link set as a group of links in a knowledge graph in an embedding space by expanding the knowledge graph with predicted and validated single links in each iteration such that a final set of links are predicted with each one being added to the set depending on previously added predicted links. In another aspect, this application also discloses automatically extracting rules from user feedback of link predictions and generating a user feedback knowledge graph from the extracted rules, which in combination with an original knowledge graph are used for the generation of the link predictions.