'I Hate the Blue Line' and Other Things Transit Systems Can Learn From Twitter
Companies have been on to "sentiment analysis" for some time now. Forget box office receipts or Rotten Tomatoes reviews. Savvy movie studios know they can now track how people really feel about a new film – literally the moment it comes out, before viewers have even left the theater – by analyzing what people say about it on Twitter. Same goes for new products, car lines or even political candidates.
So what if the same idea were applied to the service of public transit? Bus and train agencies generally gauge how riders feel about them the old-fashioned way, with surveys and focus groups. What if, instead of politely asking people if they find their morning commutes safe, sanitary and efficient, agencies tapped into the raw and unscripted assessments we all love to broadcast from our smart phones? (Case in point: I may have tweet-whined this morning from inside the Washington Metro system: "Why will it take 8 1/2 months to replace the escalators at the Dupont Metro?")
A group of researchers at Purdue suspected agencies could learn a lot about rider satisfaction by doing this (oh yeah, and all this data is free!). Craig Collins, Samiul Hasan, and Satish Ukkusuri tested the idea on the prolific tweeters who ride the Chicago 'L.' They crawled publicly available time-stamped Twitter data, including geographic location tagging, for tweets they believed came from 'L' riders, talking about the 'L.' They then weeded out all of the extraneous data. A few people, for example, turned out to be talking about "The Thin Red Line," the 15-year-old movie, not the thin Red Line, the 'L' route. The system also automatically corrected for spelling errors and style quirks (say, "I wish this train would moooooove!").
They then analyzed the content of all of these messages, culled last summer, against a sentiment algorithm with ratings for 298 positive terms and 465 negative ones, on a scale from -5 to +5. Even emoticons were included. Below are some results for what people were tweeting about last July 4. The nice tweets are represented by the blue lines, the not-so-nice ones by the red lines. (Apologies for the image quality on some of these graphs, but you should still be able to get the idea).
On the 4th of July, people seemed to get pretty ticked off about something around 10 o'clock at night. A similar pattern occurred on July 11 at about 8 a.m:
So what was going on here? This is a word cloud the researchers created, drawn from all those tweets on the 4th:
It turned out there was a big fire right around then at the intersection of Fullerton and California streets that caused huge delays on the Blue line. And here's the word cloud from the morning of the 11th:
As you might suspect, a bunch of trees fell down on the tracks, delaying the Purple, Brown and Red lines, which elicited plenty of #fail, not to mention @chicagobites.
Hasan, a Ph.D. candidate at Purdue, presented these findings Tuesday to a riveted room at the annual Transportation Research Board conference in Washington. Noticeably absent from his charts were the moments when everyone seemed to be tweeting wild praise for the Chicago Transit Authority.
"The most interesting thing we found is that transit riders do not give any positive sentiment at a particular time. They only give negative sentiment," he said. Now, this may seem depressing if you work for one of these agencies. "But that’s not very disappointing," Hasan said, "because we found that the lack of negative sentiment is basically what transit authorities should look for. If there’s no negative sentiment at any given time, that means that things are running smoothly."