Media house Delo has a 60-year tradition of excellence and unwavering trust. In this time they have become the leading and most influential company on the Slovenian media market and an important shaper of public opinion. While they originate as a print medium, Delo is known as an early adopter and innovator in the space of digitalisation. They have a huge collection of digitalised newspaper editions and photographs, which plays an important role on their journey to continue strengthening their digital capabilities. And we are glad to be taking part on this journey and helping their journalists to provide premium content by leveraging artificial intelligence.
Helping journalists to focus on creating premium content
The Struggle for Time to Work on Original Content
The role of a journalist has significantly changed over years and expectations are much broader compared to the “original” role. These days journalists are expected to do various tasks besides writing articles, such as SEO content optimisation, keyword selection, digital media creation, analytics etc. As these tasks are rather time consuming, journalists have less time to do their “main” job, i.e. researching, interviewing etc. On the other hand, readers are unrelenting and expect news to be available as soon as possible.
Thus Delo’s strategy is to automate such tasks to free up journalists’ time and enable them to focus on crafting original content and thereby even strengthen their position on the media market as being trustworthy and known for delivering premium content authored by the company’s esteemed journalists.
We joined our forces with Delo and built various AI tools that optimize SEO keywords, find related articles from archive, propose relevant photos as well as automatically generate news articles for data-rich events, such as sport events, weather forecasts, elections etc.
The Success of Automating Content Creation for NBA Game Coverage
In this case study, we focus on using AI to automatically generate news for NBA games immediately following the conclusion of an event. As sports news are mostly a summarization of statistics and post-match interviews, this AI tool significantly reduces time to prepare an article and thus frees up some time for journalists as well as provide news to readers in no time despite NBA matches usually end in the midnight hours of CET.
In this endeavour, the first step is the extraction of match scores and play-by-play data through seamless integration with the NBA API with an option to focus on key players. In our case for example, these are Slovenian players such as Dončić, Dragić, and Čančar. If available, we also gather transcripts from interviews after matches, so we can enrich news with info from various points of view.
Subsequently, we meticulously prepare structured data for integration into Large Language Models (LLMs) in a way that the output is as close as possible to the style of Delo’s journalists. A special attention is paid to the “flow” of news, which needs to be smooth and follow some specific rules.
At the end, we also generate various highlights videos for the key players based on their most attractive plays that we obtain from statistics and post-match interviews. These videos are automatically embedded in the news and referenced in the text so readers can easily find parts of their most interest.
How Technology Transformed Article Creation
We’ve harnessed AWS cloud services for automating data retrieval and processing. This choice provides a robust and scalable solution, ideal for handling large data volumes efficiently. Python, in tandem with various Natural Language Processing (NLP) packages, aids in data preprocessing, ensuring accurate and streamlined data handling from the NBA API.
A key component of our technology arsenal is OpenAI’s GPT-4, a robust Large Language Model (LLM). By utilizing past DELO articles, we’ve tailored specific prompts to fine-tune the model’s ability to craft articles in the same style as the company’s journalists.
As soon as the articles are generated, they are delivered via push notifications system as part of the Delo’s backend system and hence readily available for publishing. This also allows the system to automatically publish articles once the system is completely tested and validated.
Enhancing Reader Engagement through Customized Content
While the goal of highly automated NBA content remains in progress and there are still loads of ideas for improvements, the tool has already proved to enable journalists to prepare news article for data-rich events very quickly without any hassle. Also readers are very positive to have such news ready very early in the morning so they can enjoy reading news about their favourite sportsmen already during their morning coffee.
Such approach opens up another huge opportunity to deliver content to readers in a truly personalised way. Not just in terms of what topics they are interested in, but also how detailed articles they prefer etc. For example, some readers prefer very detailed articles on finance, while some other readers prefer broader articles on sports and just short overviews from the world of tech.
Unleashing the Potential of Automated Content Creation
This project has achieved a breakthrough in the way how journalists will be operating on a daily basis in the future, enabling them to redirect their efforts towards more intricate subjects like investigative reporting and opinion pieces, as well as Delo will be able to continue strengthening their position as a reliable and trustworthy medium. Nonetheless, the readers are provided with content quicker and with a broader range of topics.
Overall, the transformative impact of this project surpasses expectations, ushering in remarkable results and establishing a robust groundwork for future advancements. The successful integration of AI-powered solutions has revolutionized operations in the news reporting, promoting heightened efficiency, accuracy, and productivity among editorial professionals.