When talking about artificial intelligence in the newsrooms, there is too much focus on the technology and not enough on what it actually does. We want to help journalists, technophiles or technophobes, to explore this topic in an accessible way. So we are launching a new series that brings stories of your peers who work with editorial robots.
Local stories about the real estate market are relatable for most readers. After all, everyone needs to live somewhere and whether people are looking to buy or sell their home, local news coverage can become a precious source of practical news.
Interesting as they are, real estate stories are incredibly time-consuming. Mapping every neighbourhood, every property and every sale is practically impossible for larger towns — for human journalists, that is.
I’m not a tech guy. I'm a journalist.Jan Stian Vold
In 2019, Norway's regional title Bergens Tidende decided to add an automation tool, (a robot), to do some of the work human journalists do not have time to do, such as scraping public databases for property sales data and write basic hyperlocal stories. As a second largest media outlet outside of Oslo, the Schibsted-owned news outlet covers the region around Bergen, a town of some 283,000 inhabitants. That is a lot of properties to keep an eye on.
"I’m not a tech guy. I'm a journalist and I've been working in the online news business since the '90s," says Jan Stian Vold, a former editor at Bergens Tidende who is now a project lead at the title. He is working with the so-called 'Boligrobot', a piece of tech that creates automated text about the real estate market, in collaboration with the Swedish tech company United Robots.
The editorial team decided to test automation on this particular area of news coverage for two reasons. One, real estate is too wide a topic to cover properly despite the value it brings to the reader. Two, there are a lot of variables a robot can work with, and a lot of official data to sift through.
What it does
Boligrobot works on three levels. Firstly, it finds information about the real estate, like current prices, addresses, areas, changes in prices over time, the price of a square meter, and so on. This data comes from agencies and is gathered by the Norwegian authorities.
Then, it finds great aerial and panorama photos from a supplier, supplemented by images from Google street view. The aerial pictures are particularly valuable because they are taken so regularly, it shows new properties and landscapes faster than Google.
This hyperlocal data is fun for people who live in a particular postcode but hardly relevant to others. So the editors introduced a new layer: the robot does some calculations and generates property listings in a 1 km perimeter, ranking the most expensive sales in the area during the past 12 months.
"That adds relevance and that's what automated journalism should be all about," says Vold. "It’s easy to collect data and transform it to automated text. But for it to be called journalism, it needs to have a journalistic purpose and provide readers with new information they can't find elsewhere."
There are a few other features the team has introduced, like data visualisation for the whole city of Bergen or lists of top 10 sales in each municipality.
The robot can also generate headlines. Vold explained that the team created an "if this then that" structure for automated headline writing and there are some 100 headlines in the bank. The editors teach the algorithm to recognise different types of purchases, such as quickest sales or the most expensive house in a particular area.
"We want it to think like an editor," he says, adding that writing catchy headlines is one of the trickiest parts of journalistic work and one that the robot still needs to make a lot of progress on.
What is the impact?
Although the robot generated some 7,000 stories since July 2020, it is working only in addition to regular journalism.
"We still do have human reporters on the real estate market but this gives us a chance to dig deeper using the robot as a starting point," says Vold.
For instance, the team did a story about the 100 most expensive properties in Bergen and began working from robot’s data. On another occasion, the robot found that the richest person in an area just sold a property and gave a heads up to a human journalist who followed up with a more insightful story. It is also good at spotting the odd stories that generate reader interest, such as extreme prices or particular buildings.
Bergens Tidende is a subscription-based outlet so one of the main reasons to work with a robot was to engage more readers and sell more subscriptions. The bet paid off and the title sold some 600 subscriptions when marketing through the real estate automation tool. They look to sell about 1,000 subscriptions a year through this channel so they are more than halfway through. These sales contribute to around 20 - 25k subscriptions the publication sells a year.
"Journalists who cover the real estate market are fond of it because they get story leads and follow up tips," says Vold. Although there was some scepticism about readers' willingness to read articles written by a robot, early feedback suggests that the audiences are happy to. For transparency, articles written by the robot are clearly marked and it even has its own byline.
The experiment inspired further use of automated text for editorial articles that cover annual reports from companies in and around Bergen. This provides readers with useful information about local businesses and human journalists with story leads about the local economy. As far as Vold is aware, no other outlet is doing this.
This series is supported by United Robots and Utopia Analytics. Neither of them is involved in the editorial process at any stage.
United Robots AB is a Swedish technology company working in automated editorial content. The company leverages structured data to provide publishers with automatically generated content about sports, real estate, traffic, weather, local businesses and the stock market.
Utopia Analytics is a Finnish company that enables automated moderation of reader comments and cuts down the publishing delay. Inappropriate behaviour, bullying, hate speech, discrimination, sexual harassment and spam are filtered out 24/7 so teams can focus on moderation policy management.