A Match.com for Carpooling?

Several research groups are creating algorithms that make carpooling a more personalized experience

3 min read
A Match.com for Carpooling?
Photo: Ian Spanier/Getty Images

There are many reasons commuters rule out carpooling to work. Jumping into a car with a complete stranger can be terrifying and uncomfortable. It is difficult to find a driver who has the same exact schedule as you. The freedom to run to the dry cleaners or to grab take out on the way home makes having your own car feel so much simpler.

“Carpooling is not the preferred way to move people,” says Michele Berlingerio, a research scientist who studies urban dynamics at IBM Research in Ireland. There are some people who could, he says, “share a car every day, but there are potential barriers that cause people to be afraid of getting into someone else’s car.”

Some new software could change that. Two research groups, who are presenting their findings at the IEEE International Conference on Intelligent Transportation Systems this week, have devised algorithms that provide a matchmaking service to make your trip more enjoyable.

Instead of having an awkward and silent car ride with a stranger, Berlingerio’s team at the University of Pisa and IBM Research is trying to make carpooling into a social experience. They tapped Twitter data to create a carpool matchmaking model based on shared interests and compatibility of the driver and passenger. This model was compared with traditional optimized carpooling methods that aim to take as many cars off the road as possible. Of the software’s 200 users, 39 percent said they were more interested in the enjoyable, social model than in a more efficient, greener carpooling model. The engineers also found that 24 percent chose to carpool specifically for the opportunity to ride with interesting people rather than to save money and time.

The American and Italian engineers took Twitter data from residents of San Francisco and Rome and created carpooler profiles from people’s most tweeted topics, social ties, and location check-ins. A three-step algorithm coupled a driver and passenger by first analyzing tweets over the previous 30 days, ranking topics of interest, and then comparing scores of like-mindedness by looking at common followers. The algorithm then churned out a compatibility score.

Today, traveling is “going from point A to point B,” says Berlingerio, who co-authored the study. “It’s very utilitarian. By using user data, we can do a deeper analysis and more or less humanize the journey.”

Berlingerio and his team continue to fine-tune their algorithm to learn what kind of driving style a passenger likes—fast or smooth. The researchers envision trials of their social model being applied to “mini cities” like universities or large corporate campuses.

Another study, conducted by researchers at the Italian National Research Council and the Federal University of Pernambuco in Brazil, took a different approach. It matches passengers and drivers by their traveling behaviors.

Traditional carpooling systems require users to specify the locations they need to reach. These algorithms are programmed to compute the best route for one specific destination, explains the study’s lead researcher Chiara Renso, who analyzes mobility data and trajectories at the Italian National Research Council. This limits the number of carpool matches a system can make. Instead of matching people strictly based on destination, Renso and her team created a new carpool system called ComeWithMe that matches people based on destination and route flexibility. The ComeWithMe algorithm makes carpooling matches on how much a person is willing to change their destination, while preserving the activity they want to do. For example, if a driver wanted to go to Whole Foods and a flexible passenger wanted to go to Trader Joes, ComeWithMe’s algorithm would match the pair.

The algorithm is complex, says Renso, and is able it to adjust for carpoolers with rigid and flexible activities, destinations and time. The engineers first collected mobility data to better understand how people moved. They ran the algorithm on a data set of driver trajectories and found that ComeWithMe increased carpooling by more than 80 percent.

The biggest challenge Renso and her group face in their future research is measuring how much carpoolers are willing to change their original destination. “If you keep changing from one supermarket to another one, how much does a person lose interest in carpooling?” says Renso.  

People are already aware that carpooling can reduce carbon emissions and improve traffic congestion. But ride-sharers can also save up to US $1,800 per year. Still, in the United States, only nine percent of commuters carpool, according to a 2013 report from the U.S. Census Bureau containing the agency’s most recent numbers. This is down 19.7 percent from 1980.  

“I hope this can open new ideas of carpooling,” Renso says. “By looking at people’s wishes, we can improve the carpooling practice.”  

Berlingerio’s study was presented on 16 September and Renso’s study was presented on 18 September 2015 at the 18th IEEE International Conference on Intelligent Transportation Systems.

The Conversation (0)