Colin Riker
Research Mentor(s): Stefan Larson
Research Mentor School/College/Department: DryvIQ
Presentation Date: 08/03/2022
Presentation Type: Poster
Poster Number: 49
Session: Session I: 12:30 – 1:20pm
Room: League Ballroom
Authors: Colin Riker, Sara Haycox, Anastasiia Noguier, Stefan Larson
Abstract
Unstructured data is a growing occurrence in our modern data-driven society. The
prevalence of unstructured data presents issues with traditional programmatic methods for
automation and manipulation. An area of concern in unstructured data is personally identifiable
information (PII) which is an area of interest both for individuals, corporations, and governments
Thus the project chose to focus on an intersection of PII and unstructured data, street addresses.
Addresses present several fundamental challenges when attempting to identify and parse them
accurately. This comes from the complexity of postal systems and the imperfections of data
given by real people. However, there are many methods to working with unstructured data from
machine learning, pattern matching, and manual annotation. The project elected to explore how
far a pattern-matching approach could be taken. The project’s goal was to utilize a data-driven
approach to building regular expressions for unstructured data, ie. addresses. This project builds
off prior work in the form of an open-source project called the python address parser or PYAP,
written in the python programming language. The research utilizes a dataset we hand-curated and
labeled, consisting of over 2,000 documents with roughly 4,000 real addresses. Using a custom
evaluation script developed as a part of the project, issues and edge cases were identified in the
PYAP expression. From there, corrections, exceptions, and fine-tuning were made to improve
two measures of performance, recall, and precision. The expression started with 64% recall and
improved to 96% with precision starting at 84% and improving to 90%. The recall is the
percentage of all possible results returned, and precision represents the percentage of all returned
results being correct. The project’s results demonstrate that regular expressions represent a valid
and efficient method for addresses and show potential for broader applications in the space of
unstructured data.
