Connecting Crimes Algorithmically
Jan 02, 2018 — Atlanta, GA
By Alyson Powell
Late last year, Yao Xie, Harold R. and Mary Anne Nash Early Career Professor in the Stewart School of Industrial & Systems Engineering, began working with the Atlanta Police Department (APD) to test an algorithm that finds connections between crime incidents. The algorithm examines both structured data captured by 911 operators — the type of crime, and when and where it happened — and unstructured, or free text data. This type of data is gathered by police officers at the scene of the crime and includes detailed narrative descriptions from the officer, victims, and witnesses.
The tricky part for police investigators is manually analyzing thousands upon thousands of reports — including new reports that are coming in every day — to find patterns between cases, which could help solve serial crimes. It’s nearly an impossible task. Xie’s algorithm automates this process by dissecting incident reports and learning the similarities between words and common patterns in how crimes occurred. It has to be smart enough to recognize that two or more crimes could be related.
“This is an artificial intelligence way of processing police reports,” said Xie. “It’s a way of investigating cases much faster, and more effectively.”
APD provided three years of data to process, with more than 24,000 cases. The algorithm analyzed that data within hours.
“Our partnership with Georgia Tech has the potential to truly transform the speed and manner in which we currently analyze crime data,” said former APD Sergeant Frank Ruben, who is now with the department of Atlanta Information Management. “The ability this gives our investigators to proactively compare notes and identify trends will aid tremendously in furthering Chief Erika Shields’ priority of reducing violent crime through innovative technology.”
There are challenges with this method, explained Xie, including typos, grammatically incorrect sentences, and differences in how individual officers write their reports. “The reports are very different from one to the next; in fact, they’re never the same. The algorithm has to be robust enough to see errors.”
Xie is receiving financial support for her research from the Atlanta Police Foundation.