US police shooting often becomes local and national headlines, but there is no information regarding the deaths which occur. So people are turning to machine learning so that they can ensure that any shooting done by the police is not unrecorded.
It is not easy to find out how many people were killed in police operations during the police action. In a 2016 study, the US Bureau of Justice Statistics found that police records cover more than half-police-related deaths in the form of media reports. The Violent-Death Reporting System, operated by the Centers for Disease Control and Prevention, collects data from only 42 states out of 50 states and does not focus on police shootings.
According to a newsletter, the police firing had twice as a fatal record as compared to the FBI in 2015.
The most comprehensive database used by The Washington Post – considered to be fatal encounters, a website which has the list of the every person who killed by police action since 2000 in the US. It is running by a Nevada-based journalist D. Brian Burghart, in the last five years, he collected a network of online workers to collect more than 22,000 deaths.
D. Brian Burghart investigated local news reports very closely and also building a database from some other sources, which is estimated to have completed more than 90 percent.
To make this work even easier, Brendan O’Connor and his colleagues have created a system at the Massachusetts Amherst University, which automatically scrapes news reports to mention the police firing.
Brendan O’Connor said, “News articles cover a large number of details of these deaths”, By analyzing his team’s algorithm, in the news reports, trying to collect the names of those who have been killed. He trained the system for analyzing news articles from 2016, which included keywords related to some police and fatality. This algorithm separates sentences that are in the context of police firing, and then compare these sentences to Burghart’s database.
Idea was that the system gradually learned to recognize those sentences which were recently in the context of police firing, and ignored those who mentioned in historical firing or firings which were not fatal.
This system identified 57 percent of the people who were shot by the police between September and December 2016.
O’Connor is hoping to improve algorithm by putting a large range of news sites and social media data. He says “we want to extract more data from numerous sources”.
Burghart is testing the system to help the algorithm build itself its own database, if a shoot is not covered by the press, or not reported by the authorities, it can be accessed by an algorithm or researcher “It does not have any way to know what does not exist,” says Burghart.
Burghart says that machine learning will finally mean that he does not have to dig up to all news for several hours and every week so that the police-related deaths can be recorded.