2024 ADM+S Hackathon explores public service media recommender systems with the ABC
Author Natalie Campbell
Date 11 December 2024
The 2024 ADM+S Hackathon, ‘Recommender Systems for Public Service Media’, was held on 5-6 December in partnership with the ABC.
Providing an alternative to commercial forms of content curation and recommender systems is one of the challenges faced by public service media outlets in the digital era.
Hackathon participants were invited to design systems that could rank or assess content according to public service values such as social responsibility, cultural diversity, and the public interest.
“We live in a world where algorithms and automated recommendation and curation systems play a hugely important role in shaping the information that we see,” explains Prof Mark Andrejevic, ADM+S Chief Investigator at Monash University and co-organiser of the Hackathon.
Unlike commercial media, which are driven by commercial values, public media aim to serve the public good. The rise of automated systems has created challenges in ensuring that public service values are reflected in the algorithms shaping content curation and distribution.
“We’ve had four teams working hard to figure out how you might approach the challenge of developing algorithms that instead of prioritizing commercial values, prioritize what we would think of as public service values.
“The big challenge is to think about, what are those values? How do you operationalize them? How can you put them into an automated system that would recognize them? And how would you implement that in the newsroom of a public service broadcaster to support the editorial decisions that are being made.”
Leading into the two-day Hackathon, ADM+S AI James Meese was joined by industry partners Angela Ross (Research Lead), Laura Gartry (Innovation Lead) and Stuart Watt (Head of News Strategy and Innovation) from the ABC, for a panel discussion on Wednesday 4 December titled ‘What is a public service news algorithm, and why might we need one?’
This conversation, moderated by Prof Mark Andrejevic, is now available on the ADM+S podcast.
Following the Wednesday night panel, Angela Ross, Laura Gartry and Stuart Watt, as well as Saarim Saghir, Strategy manager at Google USA, engaged with the team’s ideas over the course of the Hackathon, ultimately deciding a winner to secure $5,000 in research funds for further development of their idea.
“It’s been amazing to be part of this Hackathon,” said ABC Research Lead Angela Ross.
“We’ve been blown away by the ideas and the innovative thinking, and it’s actually made us rethink some of the ways that we’re looking at the problems.”
2024 Hackathon team proposals:
‘BYO Values’
The Value Aware Ranking (VAR) Co-Pilot is an editorial toolkit which provides a scoring system to rate stories based on pre-defined and customisable public service media values.
While the ABC produces news stories with its public service values in mind, its existing recommender systems are not optimised for these values. VAR could be integrated into a back-end editorial interface and assist editorial teams to better serve their public interest function in a practical and transparent way.
VAR would help identify value gaps in certain topics that an editorial team could address with story commissioning and editing.
For consumers, VAR could help to strengthen public trust and accountability by providing transparency, explicitly linking editorial decisions to public service media values and helping consumers understand the rationale behind story prioritisation.
‘ABC News Wrap’
ABC News Wrap is an innovative news recommendation system designed to provide users with a curated list of the top 10 headlines they need to read daily, such as during their commute.
- This functionality is powered by an agentic LLM-augmented recommender system, which integrates agents with distinct priorities such as:
- ABC’s values and charter guidelines
- Users’ interests and preferences
- Community-driven trends (what other users in their proximity or demographic are reading)
‘IchiBan’
Public Service Media around the world are developing recommendation systems but often struggle to find effective ways to monitor and evaluate them. The User Behaviour Simulation for Recommender Systems proposal uses agent-based simulation methods to generate interaction data for monitoring and evaluation as an iterative process.
This solution methodology involves the following:
- LLM or manual-based generation of user profiles.
- LLM User Simulation generation.
- The generated User Simulations interact with the recommender system and give feedback for evaluation and further development of recommender systems in an iterative way.
‘Meow’
The Meow team project focused on the re-ranking section of the recommendation system pipeline, where they advocated for ABC subscribers, journalists, and editors to collaboratively operationalise public service news values. This begins with a user study, asking subscribers to allocate a public service news value of their choice to each story they engaged with on the ABC news website.
The aim of this study would be to understand how ABC subscribers’ ideas of public service news values may be similar or different to that of journalists.
–
“I really enjoyed the Hackathon because I could collaborate with people from different backgrounds to work together to solve a problem which might help the ABC,” said participant Yueqing Xuan.
“We had technical people, we also had people from design and journalism, so we could work together to make sure everything was understandable.”
The ADM+S Hackathon was organised as part of the Centre’s Research Training program for 2024.
Listen to ‘What is a public service news algorithm, and why might we need one?’
