Graph Neural Network Entity Classification Subgraphs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current entity matching systems are not accurate enough and lack the ability to explain the matching process between records and entities, with simplistic graphs not capturing real-world relationships, leading to suboptimal classification results.
Innovation Solution
Creating a training dataset comprising subgraphs of matched records with attributes, where the records are related by a subset of attributes, and using graph neural networks to classify records as belonging to an entity, providing improved learning and accuracy through the use of diverse subgraph structures.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional entity matching systems are used, then the system is simple to implement, but the accuracy of record classification is insufficient
Solution Approach 1:
The patent segments the entity matching problem into multiple subgraphs, where each subgraph represents a cluster of records matched to an entity based on specific attributes. This segmentation allows the system to handle complex relationships by breaking them down into manageable units, improving classification accuracy while maintaining reasonable system complexity through structured data organization
Solution Approach 2:
The patent introduces a new dimension by using graph neural networks that operate on multi-dimensional data structures (subgraphs with multiple attribute relationships). This transitions from traditional flat record matching to a multi-dimensional graph representation, enabling the system to capture complex real-world relationships and improve classification accuracy significantly
2Loss of information
If traditional entity matching systems are used, then the implementation is straightforward, but the ability to explain the matching process is lacking
Solution Approach 1:
The graph neural network provides feedback by generating explanations that show which attributes and relationships were most important in the classification decision. This feedback mechanism allows the system to not only classify records but also explain its reasoning, reducing information loss about the matching process while adding interpretability without excessive complexity
Solution Approach 2:
The patent uses subgraphs as intermediary structures that mediate between raw records and final classification decisions. These subgraphs serve as explanatory intermediaries, visualizing and documenting the matching process between records and entities, thereby providing transparency and explanation capability
3Reliability
If simplistic graphs are used for entity matching, then the system is easy to implement, but real-world relationships are not captured
Solution Approach 1:
The patent applies local quality by creating subgraphs with different structures based on specific attribute relationships. Each subgraph captures localized relationships between records for a particular entity, allowing the system to model real-world relationships accurately in different contexts without requiring a single complex graph structure for everything
Solution Approach 2:
The patent uses composite graph structures that combine multiple types of relationships and attributes within subgraphs. These composite structures mirror the complexity of real-world relationships by integrating multiple data sources and relationship types, improving reliability while managing complexity through modular composition
Data Source
AI summary
A computer implemented method classifies records. A number of processor units creates a training dataset comprising subgraphs of matched records matched to an entity and identifying an importance of attributes in the matched records. The matched records in a subgraph are related to each other by a subset of the attributes. The number of processor units trains a graph neural network using the training dataset. The graph neural network classifies the records as belonging to the entity.