We are happy to announce that our paper, “Beyond Top-1: Addressing Inconsistencies in Evaluating Counterfactual Explanations for Recommender Systems,” has been presented at the 35th International Joint Conference on Artificial Intelligence (IJCAI 2026) in Bremen, Germany, as part of the Sister Conferences Best Papers Track.

The paper received the Best Paper Award at ACM RecSys 2025 and investigates inconsistencies in how counterfactual explanations are evaluated for recommender systems. Our work shows that the evaluation of counterfactual explanation methods can be strongly influenced by the underlying recommender model and its performance, potentially leading to inconsistent conclusions about which explanation methods perform best.

To address this issue, we systematically study counterfactual explanation evaluation across different recommender models and performance levels and highlight the importance of moving beyond Top-1 evaluation. Our findings emphasize the need for more robust and consistent evaluation protocols for counterfactual explanations in recommender systems.

We are proud to have our research represented at IJCAI and grateful to the conference organizers and the recommender systems community for the opportunity to share and discuss our work.

We are happy to announce that our paper, "Can Third-Party Annotators Reliably Evaluate Conversational Recommender Systems?", has been accepted for presentation at the 34th ACM Conference on User Modeling, Adaptation and Personalization, taking place in Gothenburg, Sweden.

This paper investigates the reliability of using third-party crowd workers to evaluate the user experience of Conversational Recommender Systems. By analyzing over 1,000 annotations of dialogue logs, we show that while annotators can consistently judge utilitarian qualities like accuracy, they struggle to reliably assess social constructs like humanness and rapport. Our findings challenge current evaluation practices and emphasize the need for more robust, multi-rater protocols in the field.

We are thrilled to announce that our paper, "Minimal-Perturbation Counterfactuals through Guided Denoising Diffusion for Recommender Systems Explanation" has been accepted for presentation at the prestigious ACM SIGIR 2026, taking place in Melbourne, Australia.

This paper proposes DiceRec, a diffusion-based framework for generating counterfactual explanations in recommender systems. The approach learns a latent distribution of user interaction histories and guides the reverse diffusion process using gradients of the target item score, aiming to produce minimal and plausible perturbations that change the recommendation.

Stay tuned for more details about the paper!

Our team won the Best Short Paper Award at the ACM Recommender Systems Conferencefor our paper “Beyond Top-1: Addressing Inconsistencies in Evaluating Counterfactual Explanations for Recommender Systems”. 

Furthermore, Eva received the Distinguished Senior PC Member Award. We also hosted the successful BEYOND 2025 workshop, Beatriz joined the RecSys Summer School, and Michael Müller presented his PhD research at the Doctoral Consortium.

We’re excited to share that two contributions from our group have been accepted to RecSys 2025! Our short paper “Beyond Top-1: Addressing Inconsistencies in Evaluating Counterfactual Explanations for Recommender Systems” and Doctoral Symposium paper “Advancing User-Centric Evaluation and Design of Conversational Recommender Systems” will be presented at the conference.

We are also co-organizing the BEYOND 2025 workshop: Beyond Algorithms – Reclaiming the Interdisciplinary Roots of Recommender Systems.

We are thrilled to announce that our paper, "Hypergraph-based Temporal Modelling of Repeated Intent for Sequential Recommendation," has been accepted for presentation at the prestigious ACM Web Conference (WWW) 2025, taking place in Sydney, Australia.

The paper introduces HyperHawkes, a novel model that leverages hypergraphs and Hawkes Processes to capture repeated user intent in sequential recommendation systems. By integrating long-term and short-term interest modeling, HyperHawkes significantly outperforms existing state-of-the-art methods across multiple real-world datasets.

Stay tuned for more details about the paper!

The report on the Dagstuhl seminar "Evaluation Perspectives of Recommender Systems: Driving Research and Education" is out. It describes the outcomes of our Dagstuhl seminar, which brought together 41 participants from 16 countries - co-organized by Christine Bauer, Alan Said, and Eva Zangerle.

We are happy to announce that an extended version of our recent work in session-based recommendation has been accepted for publication in the prestigious ACM Transactions on Recommender Systems (TORS). This extended paper builds on our initial research, incorporating deeper insights, expanded experiments, and additional analysis that further advance the underlying model. You can find the official announcement and details about the publication here.

Marta Moscati will present our short paper on "Emotion-Based Music Recommendation from Quality Annotations and Large-Scale User-Generated Tags" co-authored by Marta Moscati, Hannah Strauß, Peer-Ole Jacobsen, Andreas Peintner, Eva Zangerle, Marcel Zentner and Markus Schedl at the 32nd ACM Conference on User Modeling, Adaptation and Personalization next week.

Our paper "Song lyrics have become simpler and more repetitive over the last five decades" was published in Nature Scientific Reports. Our study was also widely reported in the media, for instance in Scientific American, Newsweek, Science, or Der Standard.

We are happy to announce that the Special Issue on Perspectives on Recommender Systems Evaluation (PRSE) co-edited by Eva Zangerle, Christine Bauer, and Alan Said, has just appeared in ACM Transactions on Recommender Systems. It contains ten highly interesting papers that all provide different perspectives on the evaluation of recommender systems - from evaluating the user perspectives, over novel evaluation methods and frameworks, testing recommender systems, to a survey on the current state of RecSys evaluation.

Our paper The Emotion-to-Music Mapping Atlas (EMMA): A systematically organized online database of emotionally evocative music excerpts has just been published in Behavior Research Methods. In this work, we present the Emotion-to-Music Mapping Atlas (EMMA). 364 music excerpts from pop, hip/hop and classical music were rated for felt emotion using the Geneva Emotion Music Scale (GEMS). The sample comprised 517 English- and German-speaking participants and each excerpt was rated by an average of 28.76 participants. This work is part of our ongoing FWF project Humans and Recommender Systems - Towards a Mutual Understanding. 

Team RecSys will have a busy week at the ACM Recommender Systems Conference 2023 in Singapore: on Monday, Amir (paper) and Andreas (paper) will present their works in the doctoral symposium. On Tuesday, Eva (together with Christine Bauer and Alan Said) will hold the third PERSPECTIVES workshop. On Friday, Andreas will present our paper "SPARE: Shortest Path Global Item Relations for Efficient Session-based Recommendation". Furthermore,  Eva and Christine Bauer will present "Evaluating Recommender Systems: Survey and Framework" (paper), published in ACM Computing Surveys, for which they received the Women in RecSys Journal Paper of the Year award (senior category). 

We are thrilled to announce that two Doctoral Symposium papers have been accepted for presentation at the prestigious RecSys 2023 Conference in Singapore during September 2023. One paper, authored by Andreas Peintner, plans to contribute significantly to advancing Graph-based Sequential Recommender Systems. The second accepted paper, authored by Amir Reza Mohammadi, promises to make valuable contributions to the development of Explainable Graph Neural Network Recommenders. These notable contributions hold the potential to shape the future of Recommender Systems.

We are very proud that our paper "SPARE: Shortest Path Global Item Relations for Efficient Session-based Recommendation" has just been accepted as a long paper at the ACM Conference on Recommender Systems (RecSys) 2023 in Singapore. In this paper, we propose to model the multi-hop information aggregation mechanism of GNNs over multiple layers via shortest-path edges based on knowledge from the sequential recommendation domain. Additionally, we apply supervised contrastive learning to address the inherent data sparsity.