Contextual Watchlist Screening Using ML Identity Correlation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing identity matching systems for watchlists are prone to errors due to manual reviews of large volumes of data with inconsistencies and lack of integration with external data sources, leading to false positives and negatives, and require significant human intervention.
Innovation Solution
A machine learning system that utilizes natural language processing and artificial intelligence to process external data sources like social media and news articles, constructing holistic identity profiles through unsupervised semantic identity modeling and graph-based clustering, and applies machine learning models to determine watchlist candidacy with minimal explicit attribute comparisons.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual review methods are used for watchlist matching, then human judgment can assess complex cases, but the process becomes time-consuming and error-prone due to large data volumes
Solution Approach 1:
The patent segments the watchlist matching process into distinct phases: automated preprocessing of identity data, machine learning-based candidate generation, and human review of only high-risk cases. This segmentation allows the system to handle routine matching automatically while reserving human expertise for complex decisions, thereby reducing overall review time while maintaining accuracy.
Solution Approach 2:
The patent replaces manual mechanical review processes with automated machine learning models that can process large volumes of identity data rapidly. The ML models analyze name similarities, PII matching, and behavioral patterns to generate match candidates, substituting human manual review with automated computational analysis for the majority of cases.
2Adaptability or versatility
If traditional watchlist databases are used, then established data sources are available, but external valuable data from social media and news articles cannot be integrated
Solution Approach 1:
The patent creates a universal data ingestion framework that can process multiple data types from diverse sources including traditional watchlists, social media posts, news articles, and other external sources. The system uses natural language processing and entity extraction to normalize this varied data into a common format, enabling the same ML models to analyze all data types uniformly.
Solution Approach 2:
The patent introduces an intermediary layer of natural language processing and entity extraction that mediates between raw external data sources and the core matching algorithms. This intermediary layer cleans, normalizes, and structures unstructured data from social media and news articles, making them compatible with the existing watchlist matching framework without requiring fundamental system changes.
3Measurement precision
If extensive attribute comparisons are performed for identity matching, then matching accuracy can be improved, but the computational complexity and processing time increase significantly
Solution Approach 1:
The patent performs preliminary filtering and candidate generation before executing full attribute comparisons. The machine learning models first identify potential matches based on key indicators like name similarity and basic PII matching, then only perform extensive attribute comparisons on these pre-selected candidates. This preliminary action reduces the number of full comparisons needed while maintaining high matching precision.
Solution Approach 2:
The patent applies different levels of comparison depth to different attributes based on their discriminative power. Critical attributes like name and date of birth receive exhaustive comparison, while less discriminative attributes receive lighter weighting. This local quality approach optimizes computational resources by focusing detailed analysis on attributes that matter most for accurate matching.
Data Source
AI summary
Provided are a method and system for identity correlation between a transaction applicant (TA) and a watchlist entity (WE). Preexisting watchlist data and other aggregated identity data (AID) are processed to provide for comparison to a collective identity of at least the TA. Using various categorizations for the AID and the collective identity, watchlist tags are generated that can then be matched to the collective identity. As a result of the matching, a watchlist candidacy demonstrating a probability that the identity of the TA does or does not correspond to that of the WE can be generated. Various implementations of identity correlation results may be leveraged for different use cases including an automobile purchasing service, a corporate financing service, an aircraft monitoring service as well as others. Identities may comprise entities, physical objects, real estate or other items or concepts.