Entity Resolution Using Relationship Graphs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Entity resolution systems face challenges in accurately determining whether multiple data records refer to the same physical entity or distinct entities, particularly when attributes are insufficient or contain errors, leading to under-resolution or over-resolution issues.
Innovation Solution
The system determines a likeness score between entities by considering their attributes and relationships, adjusting the score based on commonly related entities, and resolves entities into a single entity when the adjusted score meets a defined threshold, thereby improving accuracy by incorporating relationship data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If entity resolution is performed based solely on attribute matching, then the process is simple and fast, but accuracy deteriorates when attributes are insufficient or contain errors
Solution Approach 1:
The patent introduces a new dimension to entity resolution by incorporating relationship data alongside traditional attribute matching. Instead of relying solely on direct attribute comparisons (one-dimensional approach), the system adds relationship information as an additional dimension (two-dimensional or higher) to the resolution process, enabling more accurate matching even when attributes are insufficient or contain errors.
Solution Approach 2:
The patent uses relationship data as an intermediary element to bridge the gap between entities with insufficient direct attribute matches. By introducing relationships as a mediating factor, the system can infer connections between entities that would otherwise remain unresolved, improving accuracy without requiring direct attribute equivalence.
2Measurement precision
If relationship data is incorporated into entity resolution, then accuracy improves for entities with insufficient attribute matches, but processing time and computational resources increase
Solution Approach 1:
The patent performs preliminary actions by pre-computing and storing relationship data before the actual entity resolution process. By preparing relationship information in advance and maintaining it in an optimized structure, the system reduces the computational burden during resolution operations, allowing accurate relationship-based matching without proportionally increasing processing time.
Solution Approach 2:
The patent replaces complex mechanical attribute-by-attribute comparison processes with more efficient relationship-based inference mechanisms. Instead of exhaustively checking all possible attribute combinations, the system substitutes this with relationship graph traversal and similarity computation, which are computationally more efficient and scale better.
3Reliability
If the likeness threshold is set low to capture more potential matches, then recall improves, but false positives increase leading to over-resolution
Solution Approach 1:
The patent implements feedback mechanisms that continuously refine the likeness threshold and matching criteria based on the results of entity resolution operations. By analyzing resolved entity pairs and their relationship data, the system adjusts threshold parameters dynamically, preventing both under-resolution and over-resolution while maintaining high reliability and precision.
Solution Approach 2:
The patent employs parameter changes by dynamically adjusting the likeness threshold and weighting factors based on the specific context, data quality, and relationship strength. Instead of using a fixed threshold, the system adapts parameters in real-time to optimize the balance between recall and precision, ensuring reliable resolution without excessive false positives.
Data Source
AI summary
Techniques are disclosed for resolving entities based on relationships to a common entity. In one embodiment, two entities are compared to determine that an entity resolution threshold is not satisfied. One or more entities commonly related to the two entities are determined. The two entities are determined to satisfy the entity resolution threshold on the basis of the one or more commonly-related entities. The two entities are then resolved into a single entity.