YouTuber Casey Neistadt makes his living on social media, but during Hurricane Harvey, he used it to save a life.

Neistadt, who has 7.7 million YouTube subscribers, lives in New York with his wife Candice; her father lives in Texas. When Neistadt learned that floodwater in his father-in-law’s Houston home had risen waist high last week, he posted a series of desperate pleas to his 1.4 million Twitter followers, including one directly tagging the Houston police department.

“VERY SERIOUS. URGENT HELP NEEDED. Candice’s 70yr old father is trapped in Houston – desperately needs rescue @houstonpolice,” Neistadt tweeted on Aug. 27. He lated added, “UPDATE #2 water has begun rising again he needs rescue urgently. he’s attempting to get on the roof. do you know anyone local?”

YouTuber Casey Neistadt posted this photo of his father-in-law’s flooded Houston neighbourhood on Twitter.

Happily, Neistadt’s father-in-law was eventually rescued. But water wasn’t the only thing overwhelming Houston during Hurricane Harvey. Over 50,000 calls poured into the city’s 911 emergency system between Aug. 26 and 27, more than six times the volume for an average one-day period.

That’s why thousands of Texans, greeted by a 911 busy signal or on hold for 45 minutes, took to Twitter instead. Twitter officials said 21.2 million tweets were posted about Harvey in a five-day span, topping tweet activity during the recent solar eclipse, all of the 2016 U.S. presidential debates and Donald Trump’s inauguration.

Phone lines at the U.S. Coast Guard (USCG) were swamped too, jammed with 1,000 calls per hour. USCG and the Houston police department posted their own tweets, asking people to dial 911 or other emergency phone numbers instead of asking for rescue assistance on social media.

Twitter as the new 911

Since millions of people across the U.S. have cut their landlines, however, the only way they can call 911 is on their mobile phones. If they keep getting a busy signal on 911, repeatedly dialing the three-digit emergency number drains their phone battery.

Twitter offers an obvious advantage over 911: if someone posts a tweet asking for emergency help, it remains visible online even hours after their cellphone has died.

If so many Texans were able to access Twitter during Harvey, why did emergency agencies tell people not to tweet rescue requests? As public safety consultant Rob Dudgeon explained to NPR in Harvey’s aftermath, those agencies simply aren’t set up to monitor urgent requests on social media.

“It’s very labor intensive to watch (social media) and because of the thousand different ways people can hashtag something or keyword something. Trying to sort out what’s relevant and what’s not and what’s actionable is very, very difficult,” said Dudgeon, former deputy director of San Francisco’s Department of Emergency Management.

The Houston police department and other local emergency services urged people not to tweet rescue requests.

Unlike phone calls, tweets can’t be automatically traced to verify the exact location of their origin or the identity of the person posting them. (Geo-tagging tweets is an optional user function on Twitter.) This makes it tough to judge the accuracy and authenticity of emergency tweets.

Fraud is such a concern that, a week before Harvey ravaged Texas, the Federal Emergency Management Agency (FEMA) posted a blog warning about fake social media activity during disasters.

Tech to analyze emergency tweets

This week, as Texas recovers from Harvey and Florida braces for Hurricane Irma, the question of how to optimize Twitter activity for emergency services during catastrophic events remains largely unanswered.

But at a recent conference in Halifax, U.S.-based researchers said they’ve created a potential solution. Called TrioVecEvent, it’s a system that uses artificial intelligence (AI) to analyze tweets and detect events like natural disasters, protests, riots and terrorist attacks as soon as they happen.

“In disaster control, it’s highly important to build a real-time disaster detector that constantly monitors a geographical region,” the scientists wrote in a research paper presented at the Knowledge Discovery and Data Mining (KDD) conference in Halifax. The event, held by the Association for Computing Machinery, focused on discoveries in data science, data mining, large scale analytics and big data.

TrioVecEvent uses algorithms to analyze the location, time and text data from geo-tagged tweets. If a large volume of tweets containing specific words and phrases suddenly appears on Twitter within a certain geographic area, the system detects “unusual activities ‘bursted’ in local areas,” according to the research paper.

How could this help in a hurricane? Although climatologists can predict the general path of a hurricane, real-time tweets from citizens can help first responders pinpoint specific damage sites almost immediately after a storm moves through an area. By correlating the locations, timing and key words of tweets right after a hurricane, emergency services could quickly determine the hardest hit neighbourhoods and what types of help they need most.

Although tweet data mining and analysis has been around for a few years, TrioVecEvent is the first solution that includes slang and abbreviations when analyzing the semantics of Twitter posts, said study co-author Jiawei Han, a computer science professor at the University of Illinois at Urbana-Champaign (UIUC).

“People use different expressions to refer to the same concept, so (traditional key word analysis) may not capture the similarity between Twitter messages too well,” Han said.

Prof. Jiawei Han

For example, TrioVecEvent includes analysis of the initials ‘KB’ in tweets about basketball star Kobe Bryant, whereas other methods usually exclude tweets containing that type of short form. TrioVecEvent’s algorithms interpret the letters ‘KB’ as a reference to Bryant if they appear in or near many tweets containing other related terms like ‘Lakers’ or ‘LA.’

Tests indicate TrioVecEvent “can achieve about 80 per cent” accuracy in detecting unfolding events vs. a 37 per cent accuracy rate with other methods, said co-author Chao Zhang, a computer science Ph D. student at UIUC.

Business applications

The location-based system has potential business applications as well.

“For tourism, if it can detect what’s happening in a city at different locations in different time periods, it can recommend interesting things for a visitor according to their preferences,” Zhang said.

Han said other use cases could involve advertising and marketing.

“Maybe around Fifth Avenue in New York City, say a lot of people are posting shopping related stuff on social media. On the business end, a retailer can know the activities those people are interested in at different locations and at different times. So they can design their advertising strategies according to that information or knowledge,” Han said.

Share on LinkedIn Share with Google+
More Articles