Djinn — Data Journalism Interface for Newsgathering and Notifications
2024 Finalist

Djinn — Data Journalism Interface for Newsgathering and Notifications

iTromsø

Tromsø, Norway

Category Artificial Intelligence

Media associated with this campaign

Overview of this campaign

iTromsø started out wanting to utilize machine learning and AI technology to help our journalists reduce the amount of time spent going through archives and documents, and to find stories we otherwise wouldn’t.

At the same time we saw a demand for strengthening our coverage of the building, zoning and urban development beat.

Most Norwegians today invest a substantial amount of their savings in their own home, and the decisions made by developers and local government can have far reaching impact on communities, neighborhoods as well as individuals.

We also see that this news category is one of our most read and one of the categories that generate the most subscription sales, but at the same time one of the categories with the least produced stories.

This is due to how challenging the beat is. To find the relevant stories for our readers, we must check the municipal mail journals every day, which often includes several hundred documents and entries without any substantial information about the contents of the documents, which means we manually have to check every document, which again is an error-prone process where we miss out on stories often. This made the urban development beat an ideal testing candidate for AI solutions.

By improving our coverage and picking up more of the relevant stories within the domain, we want to keep our readers informed about what’s going on and by extension increase participation in the local democratic processes that shape our city.


Results for this campaign

We’ve now developed the Djinn platform in collaboration with IBM and their partner Visito.

Djinn downloads all documents from the municipal archives every day. The documents are then sent through a pipeline where a central and a local AI-model ranks them based on newsworthiness. The documents are then summarized by an LLM and entity extraction and outlier detection is performed. The documents are then displayed in a list view within the Djinn platform, so journalists get the highest ranked documents listed at the top every day, and with a glance can make judgements on the actual news value using the summaries and the relevant entities displayed for each document.

Djinn also sends email notifications with the top-ranked possible stories every morning.

As journalists use the platform, they give feedback on the rankings and further improve the models.

What we’ve seen so far is that the research time is reduced by over 80 percent. It used to take about an hour for an experienced journalist to go through the archives for each municipality, now it takes ten minutes.

We’ve shortened the time-to-publish substantially, we’re picking up stories we wouldn’t otherwise, while at the same making the beat more accessible to new and inexperienced journalists. This has led to better coverage of the domain for our readers.

We’re now scaling up the Djinn platform to four of the largest papers within Polaris Media, iTromsø’s owner. By the end of February this year we hope to have rolled out the platform to 36 of the papers in the corporation. The solution will cover about 150 municipalities in Norway.

Going forward we’d like to explore other domains and document types within the platform, as Djinn itself, broadly speaking, is a news discovery system that can be utilized on any publicly available text document.


Contact

To contact a company representative about this campaign, click here for the INMA Member Directory

By continuing to browse or by clicking “ACCEPT,” you agree to the storing of cookies on your device to enhance your site experience. To learn more about how we use cookies, please see our privacy policy.
x

I ACCEPT