Tutorials

• Data Access for Recommender Systems Research: leveraging the EU’s Digital Services Act by João Vinagre, Lorenzo Porcaro, Silvia Merisio, Erasmo Purificato and Emilia Gomez

  Abstract

  The Digital Services Act (DSA) regulates how online platforms and search engines may operate within the EU. Platforms and search engines with more than 45 million monthly active recipients are designated Very Large Online Platforms (VLOPs) and Very Large Online Search Engines (VLOSEs) and are subjected to specific obligations, given their potential societal impact. These obligations are related to the systemic risks identified in the DSA which fall under four categories: a) the dissemination of illegal content; b) risks to the exercise of fundamental rights; c) negative effects on civic discourse and electoral processes, as well as public security; and d) negative effects related to ender-based violence, protection of minors and public health, and negative effects to physical or mental well-being of individuals.

  Given the lack of consolidated scientific knowledge about the impacts of VLOPs and VLOSEs on society the DSA invites researchers from anywhere in the world to study and contribute to the understanding of systemic risks. Researchers can request data from VLOPs and VLOSES if their research is clearly connected to systemic risks in the EU and satisfies a set of other conditions described in the regulation. With this, the DSA provides an unprecedented legal mechanism for addressing a wide range of research questions that have previously largely been left unanswered due to the unavailability of data. The goal of this to tutorial is to encourage researchers to submit DSA data access requests and to equip them with the necessary knowledge to do so.

• Multi-armed bandits in the wild by Kim Falk

  Abstract

  Recommender systems present distinct challenges compared to traditional machine learning tasks; making recommendations is not like predicting if it will rain tomorrow. Making recommendations requires making decisions in the face of uncertainty.

  A key approach to tackling this problem is through multi-armed bandit algorithms, which enable systems to strike a balance between exploration and exploitation. While conceptually straightforward, implementing bandits in production systems is notoriously challenging.

  This tutorial introduces some of the most commonly used bandit algorithms, guiding participants through the transition from zero to production. We will address common hurdles, including the lack of instant feedback in many systems, the risks of exploring too broadly, and the difficulty of tuning the exploration-exploitation trade-off to fit specific use cases. To provide hands-on experience, we will use a Jupyter notebook to simulate a bandit behaviour and demonstrate its performance in a simulated production environment, highlighting best practices and illustrative pitfalls. By the end of the tutorial, participants will have acquired the theoretical foundations and practical tools necessary for building and deploying bandits in the context of recommender systems.

• A Hands-on Dive Into Quantum Computing for Recommender Systems by Maurizio Ferrari Dacrema and Paolo Cremonesi

  Abstract

  The field of Quantum Computing (QC) has gained significant popularity in recent years due to its potential to accelerate computationally intensive tasks. However, QC is a highly complex field that presents a significant entry barrier for newcomers. This tutorial aims to provide an accessible introduction to QC for an audience unfamiliar with the technology, followed by a demonstration of how the Quantum Annealing (QA) paradigm can be applied to solve practical problems in Recommender Systems.

  The session begins with a high-level overview of QC fundamentals, introducing key concepts such as the Quantum Circuit and Quantum Adiabatic models. Participants will then learn how to frame classical optimization problems as Quadratic Unconstrained Binary Optimization (QUBO) models, a necessary step for Quantum Annealing. The tutorial will explore practical applications of QC in Recommender Systems, focusing on tasks like feature selection, community detection, and clustering. The session concludes with hands-on coding exercises, where attendees will work directly with a quantum computer through cloud platforms, gaining practical experience in applying QC to realistic problems.

• Agentic LLM for Recommender Systems by Chengkai Huang, Junda Wu, Tong Yu, Julian McAuley and Lina Yao

  Abstract

  Data-driven recommender systems have achieved remarkable accuracy via deep learning, yet they typically operate as static predictors that struggle to adapt in real time or anticipate evolving user needs. Recent breakthroughs in large language models (LLMs) introduce planning, reasoning, and dynamic memory capabilities that can elevate recommenders from reactive tools to autonomous agents.

  In this tutorial, we offer a comprehensive overview of agentic recommender systems powered by LLMs. We begin by defining the core agentic capabilities such as context interpretation through chain-of-thought prompting, dynamic memory for sequential profile updates, and policy-driven action selection, and contrast these with traditional collaborative filtering and neural approaches. Next, we dive into the design of modular system components, illustrating how to integrate LLM-based reasoning with user modeling pipelines. Through real-world case studies and comparative analyses, participants will learn practical implementation strategies. We then address critical challenges such as explainability, bias mitigation, safety considerations and lifelong personalization, and survey emerging solutions. Finally, we outline open research directions, evaluation frameworks and benchmark design principles to guide future innovations. Attendees will leave equipped with methodological insights and practical guidance for building the next generation of intelligent, agentic recommender systems.

• Multi Agentic Recommender Systems: Foundations, Design Patterns, and E-Commerce Applications by Reza Yousefi Maragheh, Yashar Deldjoo, Chi Wang, Jason Cho and Derek Cheng

  Abstract

  This tutorial presents a roadmap for building large‑scale, multi‑agent recommender systems powered by large language models (LLMs). Drawing on deployments at industrial RecSys teams and on contributions to open‑source frameworks, we synthesise the “alphabets” of agentic design: memory moderation and retrieval, function calling and tool use, context protocols, and reasoning‑load balancing. Participants will learn how these components enable dynamic, multi‑step recommendation workflows that surpass conventional one‑shot approaches in tasks such as conversational recommendation, context‑aware autonomy, evaluation, user‑behaviour simulation and explanation generation.

  The three‑hour, half‑day program combines concise conceptual modules with a live, hands‑on notebook in which attendees build and debug a personalised “birthday‑planner” agent pipeline. Throughout, we highlight best practices for scalability, cost control, and compliance, and discuss emerging research frontiers including advanced memory, multi‑agent responsible orchestration.

  The tutorial targets researchers, doctoral students and industry engineers who possess working knowledge of recommender systems or LLMs and wish to translate recent research breakthroughs into reliable production systems. Attendees will leave with actionable design patterns and an informed view of how agentic architectures are reshaping the future of recommendation systems across industries and academic research. For more info please visit the website of the tutorial.

• Standard Practices for Data Processing and Multimodal Feature Extraction in Recommendation with DataRec and Ducho by Alberto Carlo Maria Mancino, Matteo Attimonelli, Angela Di Fazio, Daniele Malitesta

  Abstract

  The common recommendation pipeline involves many stages, from data preprocessing and model’s training to recommendation performance evaluation. Unlike other pipeline stages, dataset preprocessing has generally received less attention in the literature, leading to a lack of standardization, comparability, and reproducibility of recommendation systems. To this end, we present our tutorial on “Standard Practices for Data Processing and Multimodal Feature Extraction in Recommendation with DataRec and Ducho (D&D4Rec)”. To begin with, the tutorial will shed light on established strategies for data preprocessing in recommendation, aiming to provide a formal and comprehensive taxonomy of standard practices. Such theoretical notions will be complemented with a hands-on session on DataRec, a recently proposed framework for the standardization of data preprocessing in recommendation. Then, the tutorial will tailor a specific but popular scenario in recommendation, namely, multimodal recommendation. After presenting common procedures to extract and process multimodal information enhancing the user-item recommendation data, the tutorial will feature a second hands-on session on Ducho, a popular framework for multimodal feature extraction in recommendation. More details about the tutorial will be continuously updated and made available at: https://sites.google.com/view/dd4rec-tutorial

• Recent Advances in Generative Conversational Recommender Systems by Thomas Elmar Kolb, Ahmadou Wagne, Ashmi Banerjee, Fatemeh Nazary, Julia Neidhardt, Yashar Deldjoo

  Abstract

  Conversational recommender systems (CRSs) are increasingly vital for delivering multi-turn, context-aware recommendations. This tutorial provides a concise yet comprehensive exploration of modern generative CRSs, highlighting recent advances in generative AI—such as breakthroughs in large language models and neural generation pipelines, that enhance dialogue management, user modeling, and response generation. In addition, the tutorial addresses core challenges, including data acquisition, multi-turn personalization, and evaluation issues, such as controlling hallucinations, accounting for social factors, and managing ethical considerations, while also discussing emerging risks and novel solutions. Ultimately, participants will be equipped with actionable insights and practical tools for building new conversational recommender systems powered by generative models.

Tutorials Chairs

• Bart Knijnenburg, Clemson University, USA
• Elisabeth Lex, Graz University of Technology, Austria
• Fedelucio Narducci. Politecnico di Bari, Italy

E-mail: