Last week, the ISG team—Prof. Joeran Beel, Moritz Baumgart, Lukas Wegmeth, Tobias Vente, and Philipp Meister—attended the ACM RecSys 2025 conference held at the O2 Universum Convention Center in Prague, Czechia. As the premier venue for recommender systems research, ACM RecSys gathers leading academic and industry groups to present state-of-the-art techniques, evaluation methods, and real-world deployments.
Volunteering
During the event, Lukas, Moritz, and Philipp served as Student Volunteers. They assisted in coordinating sessions, managing registration, and supporting logistics between talks. This role offered a close look at how large conferences are run, and provided opportunities to interact with authors, attendees, and conference organizers in more informal settings.
Themes at RecSys & Prague impressions
This year’s RecSys program focused on LLMs and Transformer architectures, exploring new ways these models can enhance recommendations. The conference also featured developments in hybrid recommendation methods, combining different approaches to improve performance. Another key theme was fairness, interpretability, and reproducibility, highlighting the community’s ongoing efforts to make recommendation research more robust and transparent.
In addition to attending sessions and workshops, we explored Prague whenever time permitted. Visits to Prague Castle, the Charles Bridge, and the winding streets of the Old Town offered a chance to experience the city’s historic atmosphere.
Tobias presents APS Explorer
A highlight for our group was the poster by Tobias Vente: “Navigating Algorithm Performance Spaces for Informed Dataset Selection”, co-authored with Michael Heep, Abdullah Abbas, Theodor Sperle, Jöran Beel, and Bart Goethals.
About APS Explorer
In offline recommender-systems research, dataset choice can strongly influence experimental outcomes. Yet, as discussed in our poster paper blog post, many studies still rely on a few standard datasets without clear justification. APS Explorer addresses this by offering an interactive, web-based tool for exploring Algorithm Performance Spaces (APS)—a framework that relates datasets and algorithms based on their performance patterns.
The tool provides intuitive visualizations, including PCA-based clustering of datasets, meta-feature comparisons, and pairwise algorithm performance views. These features help researchers select datasets more systematically and better understand performance relationships across algorithms.
RecSys 2025 was a valuable opportunity for our team to connect with researchers from around the world, and gain insight into emerging trends in recommender systems. The discussions and feedback we received will help guide our ongoing work in dataset selection and algorithm evaluation.
0 Comments