Graph-Based PII Coreference Resolution for Privacy Compliance
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current content management systems lack effective techniques to confidently and securely associate personally identifiable information (PII) with specific individuals, especially when the context does not explicitly identify the person, and struggle to manage PII embedded across large volumes of content objects.
Innovation Solution
The implementation of graph-oriented data structures and natural language processing techniques for coreference resolution allows for the identification and association of PII with individuals by processing a corpus of content objects, recognizing aliases, and correlating PII through linguistic analysis, thereby reducing memory and processing requirements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing PII identification approaches are used, then PII instances can be detected in content objects, but the ability to correlate PII to specific people is limited and confidence is insufficient
Solution Approach 1:
The patent introduces an intermediary entity graph structure that mediates between PII detection and person identification. The graph uses coreference resolution as an intermediary mechanism to bridge the gap between detected PII instances and actual persons, enabling confident correlation through linguistic analysis of contextual relationships rather than direct association
Solution Approach 2:
The patent replaces traditional mechanical PII matching approaches with natural language processing techniques. Instead of relying on explicit mechanical links between PII and persons, the system uses linguistic coreference resolution to infer relationships through text analysis, achieving higher confidence in associations
2Quantity of substance
If content management systems store large volumes of content objects with embedded PII, then comprehensive content management is achieved, but memory and processing requirements increase significantly
Solution Approach 1:
The patent extracts only the necessary PII-related information from the full content objects and represents it in a compact graph structure. By taking out only the essential PII entities and their relationships rather than processing entire content objects, the system reduces memory and processing requirements while maintaining comprehensive PII management capability
Solution Approach 2:
The patent segments the PII management task into discrete graph operations. Instead of processing all content objects uniformly, the system divides the problem into: (1) PII detection in content objects, (2) coreference resolution to identify persons, and (3) graph construction to store relationships. This segmentation enables efficient processing of large volumes of content
Data Source
AI summary
Various corpora of content objects and other information sources beyond the corpora of content objects are processed to identify personally identifiable information (PII). PII that is associated with a named person is codified into a first portion of a graph-oriented data structure. Also, PII that is associated with an alias that might refer to a named person is codified into a second portion of the graph-oriented data structure. A determination is made that the alias that might refer to a named person is indeed an alias that is a coreference to the named person. Based on that determination then, since the first portion of the graph and the second portion of the graph refer to the same person, then the PII of the second portion of the graph can be deemed to be PII of the same person. PII in common and/or language processing can be used to establish coreferences.