About POPROX
What is POPROX?
- POPROX is an experimental platform for recommendation research. It's a cloud-based system where researchers can test new recommendation algorithms and interfaces.
- POPROX News uses a panel of US adult readers who receive a daily email newsletter of AP news.
- The platform handles consent, payments, and surveys to gather user feedback.
- It also tracks user behavior (like clicks) and provides standardized evaluation methods.
- A community advisory board, including researchers and news media, oversees the system.
- POPROX is currently a news digest. In the future we intend to develop a news app and support recommendation in domains other than news.
Why POPROX?
Human-centered evaluation of AI systems (HCAI) requires observing and interacting with actual humans using those systems. For a large subset of AI researchers, this is an unfamiliar, slow, and costly process that is prone to significant errors. We developed POPROX, in part, to encourage and help AI researchers to test their algorithms on actual users.
Who is POPROX?
POPROX is collaboration between five universities and other partners. See our People and Partners page for more information.
The POPROX Ethos
Open science promotes transparency, collaboration, and openness in research practices. We want to encourage the practice of open science by asking experimenters to register their designs before collecting data, offering standardized evaluation metrics and software, and making it possible to obtain data from experiments that have already been run on the system.
Why news?
- News recommendation presents significant technical and societal challenges.
- News data is complex, involving text, images, and frequent updates ("item cold start").
- Improving news delivery and public information are crucial societal goals.
- POPROX provides opportunities for collaboration with social scientists and journalists.
- The rise of large language and image models makes POPROX a valuable testing ground for these cutting-edge technologies.