Disruptions are made all the more infuriating when you have to grapple for information about what, exactly, has gone wrong and when you can expect to start moving again. One recourse is to start scanning Twitter for answers, but that can feel a little like listening to murmurs down a corridor. Often you’re left confused, twiddling your thumbs until the train lurches back into life.
Ticket company Trainline wants to change all of that, offering its customers a way to cut through the noise with tailored disruption updates. “We wanted to make it easier for our users to receive relevant contextual disruption information at a point in time where they might need it,” says Fergus Weldon, the company’s director of data science.
The in-app system makes use of machine learning techniques, but the core of it is relatively simple: Trainline has built a system that takes in tweets sent by the various rail operators, then automatically pulls relevant information from the messages and uses this to create its own sense of what is happening.
READ NEXT: How biometric tickets, sensors and drone repairs could spell the end of train delays and strikes
“We created a context scoring system,” explains Weldon. “We have three classifications for the severity of the message. From the [operators’] messages we also parse out the station, as well as directions of the station; say, going from A to B, or from B to A.
“Using all of that information, we overlay that with a real-time view of where the trains are in the network. So we can work out individual trains that are affected, and can pass that information to the people travelling on those trains.”
While the information will initially only be available through Trainline’s voice app, the eventual idea is that notifications will prod travellers with disruption info on their journey – as long as its pertinent. “In today’s day and age, we all receive enough notifications, so we’re not trying to send you information that you don’t need to know about,” Weldon reassures.
“If you’re a regular commuter, whether you’re in the South East, Edinburgh or Manchester, then you will only get disruptions that are only relevant during the particular times of day you travel.”
AI travel
In 2017, Trainline launched its price prediction service; using billions of data points to predict when ticket prices are liable to rise and fall. Around the same time, the company launched BusyBot, an AI bot that crowdsources 26,000 passengers’ messages every day about where they’re sitting and whether or not the seat next to them is free, then uses this to point customers to the less-populated carriages.
Disruption notifications are the latest part in the wider move by Trainline to leverage AI in a way that can help it feed useful information to passengers. The company doesn’t only want to sell tickets, it seems, but aims to see its customers through the journey as well. “They’re getting the best price,” Weldon enthuses. “They feel like they’re being informed if there are issues on travel. In the future we want to get to a point where we can help them understand if they’re going to be affected, even if they don’t know it yet.”
Building a system that pulls information from disparate Twitter feeds isn’t as simple as it first sounds: “The hard part was, as with any freeform natural language, it’s unstructured. You can’t always predict how humans will type a particular message. So we’ve had a lot of fun trying to build models that accurately predict the structures of sentences.”
Pulling information from official train operators should hopefully stop disinformation from slipping into Trainline’s system, but the notifications still rely on third-party information being posted to Twitter. If, for whatever reason, the likes of Great Western or Virgin Trains were to stop using the platform as much as they currently do, that would leave Trainline without a paddle.
Regardless, Trainline is making an effort to hoist pre-existing information into its platform, and if it works it could help make the company’s app a handy travel partner.
Trainline’s disruption notifications will be coming to beta “soon”, according to the company.
Disclaimer: Some pages on this site may include an affiliate link. This does not effect our editorial in any way.
