Entity Resolution via Relationship Graph Aggregation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing information search methods are incompatible with large datasets, such as those found on the Internet and cloud servers, requiring prohibitive amounts of indexing to capture and build relationships among vast amounts of data, especially when dealing with entities like e-mail addresses, phone numbers, and user identifying information.
Innovation Solution
A relationship graph system that infers relationships between entities by analyzing observations, representing them as nodes and edges, assigning scores based on metadata, and combining edges to generate aggregate relationships, while also considering confidence levels and noise levels to determine valid connections.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional indexing methods are used to build relationships among large datasets, then relationship accuracy can be maintained, but computational resources and time required become prohibitive
Solution Approach 1:
The patent segments the large dataset into multiple partitions or shards distributed across different nodes. Each node processes local data independently to build local relationship graphs, avoiding the need to index entire datasets centrally. This segmentation enables parallel processing while maintaining relationship accuracy through distributed computation.
Solution Approach 2:
The patent introduces intermediary summary statistics or aggregated relationship features that mediate between raw data and final relationship inference. These intermediaries pre-compute and store essential relationship patterns, allowing downstream queries to leverage pre-processed information without re-indexing entire datasets, thus improving computational efficiency.
2Measurement precision
If more data is collected to improve relationship inference accuracy, then confidence in relationships increases, but data processing time and storage requirements increase
Solution Approach 1:
The patent performs preliminary actions by pre-computing and storing summary statistics, aggregated features, and pre-processed relationship indicators during data ingestion. When relationship inference is needed, the system leverages these pre-computed intermediaries rather than processing raw data from scratch, significantly reducing processing time while maintaining confidence through accumulated evidence.
Solution Approach 2:
The patent changes parameters by transforming raw data into aggregated features and summary statistics with different statistical properties. This transformation reduces data dimensionality and complexity while preserving essential relationship signals, enabling faster processing without sacrificing inference accuracy through the use of statistically robust aggregated measures.
3Difficulty of detecting and measuring
If complex indexing structures are built to capture relationships among entities, then relationship detection capability improves, but system complexity and maintenance difficulty increase
Solution Approach 1:
The patent extracts essential relationship signals from complex data by identifying and isolating key features and patterns that define entity relationships. Instead of building comprehensive indexing structures to capture all possible relationships, the system extracts and processes only the most salient relationship indicators, reducing system complexity while maintaining effective relationship detection capability.
Data Source
AI summary
A relationship graph system disclosed herein provides a system and method for inferring relationships between various entities based on analysis of a plurality of observations including elements related to two different entities. The method comprises representing each of the plurality of observations using a relationship edge between two nodes, each of the two nodes representing an element related to one of the two different entities, assigning a score to each relationship edge based on metadata related to the observation represented by the relationship edge, combining a plurality of relationship edges between the elements related to the two different entities to generate an aggregate relationship edge between the two different entities, and inferring a relationship between the two different entities based on length of the aggregate relationship edge between the two different entities.