Twitter Makes It Real

Twitter Could Help Police Predict Where Crimes Will Be Committed

Be careful what you hashtag.
Your friendly neighborhood Time Police. (Photo via Eneas De Troya)

Your friendly neighborhood Time Police. (Photo via Eneas De Troya)

A research team has developed a method to predict crime patterns using tweets, and according to Motherboard, the NYPD has already shown interest in the potential.

The study from The University of Virginia, called “Predicting Crime Using Twitter and Kernel Density Estimation,” used geotagged tweets to find hot spots for crime. The team ran their trial experiment in Chicago, collecting both a record of all crimes committed between January 1st and March 31st 2013, and all of the tweets geotagged within Chicago during that time.

The study is just the next step in prediction techniques police have been using for years. Kernel density estimation (KDE), which looks at past crime patterns and determines hotspots where crime is likely to occur again, was the inspiration for the models built study.

KDE is useful for helping officers know where they should keep an eye out, but these models are vague outlines, like predicting the weather on last year’s forecasts — you might know that it usually rains in April, but that doesn’t tell you exactly which days to carry an umbrella. Adding tweets as another layer of data adds a live dimension that can improve crime mapping.

The analytics engine is largely driven by looking at keywords: criminal damage often occurs around sporting events, so “Blackhawks” and “Bulls” can put up red flags, where “lounge” and “studios” help geomap prostitution (“Ukrainian” made the list too, for reasons we’d rather not speculate on).

“For 19 of the 25 crime types we studied, the addition of Twitter data improves crime prediction performance versus a standard approach based on kernel density estimation,” the study said.

Twitter was especially helpful in predicting stalking, gambling and criminal damage, but was actually counterintuitive when trying to predict arson, kidnapping and intimidation. The study says it’s tough to figure out why certain types of crime respond better to Twitter analysis.

As Motherboard points out, we’re a very long way off from not just adapting this program, but even having a technological framework in the NYPD that would allow for live tweet monitoring across the five borough. They’re going to at least have to figure out how this “hashtag” thing works before they start using twitter to analytically profile you.

Follow Jack Smith IV on Twitter or via RSS. jsmith@observer.com