Case Study: Djinn, an AI-powered Data Journalism Interface


Published August 20, 2024

AI in the Newsroom is a series of case studies from the Online News Association (ONA) that highlight specific ways journalists are building and using AI tools.

iTromsø, a small Norwegian newsroom that is part of the Polaris Media group, has developed an AI-powered data journalism interface called “Djinn” to enhance its newsgathering and notifications process. The tool is designed to assist journalists in finding, analyzing and summarizing news stories efficiently.

Opportunity

Journalists often face challenges in efficiently sifting through vast amounts of data to find relevant news stories, particularly in small newsrooms with limited resources. iTromsø, a 20-perso

n Norwegian newsroom, recognized this challenge as their journalists spent 2-3 hours daily searching through municipal archives and web portals, often missing important stories due to time constraints and the limitations of relying solely on document titles.

The newsroom aimed to create a tool that would streamline their research process, ease the time-consuming tasks of data processing and document analysis and ultimately help journalists uncover exclusive news stories and trends.

Solution

iTromsø developed Djinn—an AI-powered tool designed to streamline research and news gathering. The development process began with a full-day workshop involving journalists, developers and AI specialists to understand the specific difficulties faced by the newsroom. They discussed potential solutions and aligned them with their goals.

The team that collaborated to create Djinn consisted of approximately 15 people, including five from the newsroom, specialists from IBM and a local company called VC2. The prototype was developed in about two months, involving around 1,000 hours. The tool incorporates a data pipeline that collects data from various municipal sources using custom-built scrapers and APIs.

Key components of Djinn

  • Data Scraping and Collection: Djinn uses custom-built scrapers and APIs to gather data from various municipal archives, collecting over 12,000 PDF documents monthly.
  • AI-powered Document Analysis: The tool uses machine learning to rank documents based on relevance and generates auto-summaries as well as extracts key information, enabling journalists to quickly identify relevant content and make informed decisions about which documents to investigate further. It uses advanced AI models, such as BERT for Norwegian language processing and Meta’s Llama for cost-effective summarization, to analyze and process the collected documents.
  • Investigative Lead Generation: Djinn’s AI algorithms detect patterns and trends within the data, recommending potential issues and stories for journalists to pursue. By identifying key individuals and exposing hidden connections, Djinn provides journalists with valuable starting points for their investigations, helping them uncover exclusive stories that might otherwise go unnoticed.
  • Customizable User Experience: Journalists are able to create custom searches and follow specific topics of interest. The tool’s tailored approach ensures that journalists can quickly access the most relevant information for their work, saving time and effort in navigating through vast amounts of data.

This diagram illustrates the key components and steps involved in the Djinn AI-powered workflow.

Outcomes so far

The implementation of Djinn has led to several positive outcomes:

  • Journalists now spend only 10 minutes per day reviewing potential stories, compared to 2-3 hours previously.
  • The newsroom has discovered many stories that were previously missed, improving the quality and depth of their reporting.
  • The tool has helped shift the focus from “what” is happening to “why” it is happening, enhancing the quality of news stories.
  • The newspaper has achieved a record-high number of subscriptions, with a 15% increase in the past year.

The tool’s success has prompted the Polaris Media group to adopt Djinn across 35 of its newspapers, with a monthly operating cost of $5,000.

Lessons learned

  • Involving journalists throughout the development process ensures the tool meets their needs and fosters a sense of ownership.
  • Training the AI algorithm using real-time feedback from journalists in their work environment is more effective than isolating the training process.
  • Collaborating with external partners, such as IBM and local companies, can provide valuable expertise and resources for small newsrooms.
  • Investing in AI-driven tools can lead to significant improvements in efficiency, content quality, and business outcomes, such as increased subscriptions.

This resource is part of the AI in the Newsroom series. Read other case studies you might have missed:

Related Topics
Author
Nikita Roy
ICFJ Knight Fellow and Newsroom Robots Podcast Host