Accepted tutorials
Presented by: Kurt Cutajar, Jas Kandola, Anjun Hu and Yashar Deldjoo
Abstract: Recommendation and retrieval systems are increasingly built by composition, with pipelines of LLM-powered agents that retrieve, reason, and act, replacing standalone models. This shift inherits long-standing risks while introducing new ones, yet the resulting threats to robustness and safety, and the strategies to contain them, remain underexplored. Focusing on multi-agent recommender systems, this tutorial organizes these threats along three tiers, from drift through misalignment to compromise, and separates risks that composition merely amplifies from those that emerge only once agents interact. Because agents that each pass their own safety checks can still interact in unsafe ways, we also show how to scope evaluation to the level at which each failure surfaces: the individual agent, the interaction, or the composed system. Interactive walkthroughs of simulated attacks illustrate how these failures play out end-to-end across recommender and retrieval pipelines alike.
Abstract: Recommender systems are essential to real-world online services due to their substantial business value. Increasingly, large language models (LLMs) have been adopted for recommendation tasks, owing to their strong reasoning capabilities and their effectiveness in handling cold-start items. A large body of work has sought to adapt retrieval-augmented generation (RAG) for LLM-based recommender systems by retrieving information from external databases and integrating it into the LLM’s input to improve recommendation quality. Despite the rapid growth of the field, existing tutorials lack a clear taxonomy and a dedicated review of RAG methods for LLM-based recommender systems. To fill this gap, we present the first tutorial on this topic. First, we discuss the types of information used in LLM-based recommender systems, including target-user information and external databases for retrieval. Second, we present the design space of RAG for LLM-based recommender systems, outlining the key design choices at each stage of the pipeline: (1) retrieval, (2) augmentation, and (3) generation. Lastly, we introduce open questions for future research on this topic.
Presented by: Stefano Marchesin, Gianmaria Silvello and Omar Alonso
Abstract: Knowledge Graphs (KGs) are foundational infrastructure for search, question answering, and entity-centric applications, on the open web and within domain-specific platforms. Their lifecycle, from construction over heterogeneous sources to evaluation at scale and use in downstream tasks, raises a coupled set of data management problems that shape the reliability and utility of every system built on top of them. This tutorial offers a high-level view of KG data quality at scale, organized along KG lifecycle stages. We first frame KG construction and the open challenges that arise when facts are extracted from noisy, evolving sources. We then turn to evaluation, covering efficient quality estimation, as well as the role of Large Language Models (LLMs) as validators and auxiliary signals. We close with downstream applications, including entity-oriented search and scalable defect detection, which translate evaluation outcomes into operational decisions.
Presented by: Fabio Massimo Zanzotto, Elena Sofia Ruzzetti and Leonardo Ranaldi
Abstract: Transformer-based Large Language Models (Tb-LLMs) demonstrate extraordinary capabilities and, thus, change the approach of conducting research even in Knowledge Representation. Building on existing research that neural networks (NNs) on discrete symbols operate as holistic reasoners over “classical” symbolic approaches, this tutorial aims to present the dominant paradigm of Tb-LLMs according to the idea that these models can be understood with neuro-symbolic, ontology-grounded approaches and controlled through mechanistic interpretable or explainable KG reasoning methods. Offering a profound understanding of the inner workings of these models, the tutorial will help practitioners build novel, transparent architectures to model symbolic knowledge in NN.
Presented by: Joao Gama and Arijit Ukil
Abstract: Generative AI is increasingly embedded in information retrieval, knowledge management, and recommender systems, yet its adoption is constrained by finite compute, memory, latency, energy, and data-governance budgets. This tutorial presents a resource-aware and sovereign perspective on generative AI for scalable knowledge access, framing efficiency as a multi-objective optimization over accuracy, latency, and cost. We structure the tutorial around: intelligence building, which addresses training and adaptation under explicit resource budgets, and intelligence delivery, which focuses on serving-time inference under strict tail constraints (e.g., P99 latency, peak memory) typical of large-scale retrieval and decision-support pipelines. Core techniques, including parameter-efficient fine-tuning, sparsity and pruning, quantization, distillation, mixture-of-experts routing, and speculative decoding, are analyzed as controlled approximations that trade utility and predictive uncertainty. The tutorial equips participants with methods and deployable design patterns for building cost-efficient, privacy-preserving generative AI systems aligned with CIKM’s core themes of scalable, trustworthy knowledge access under real constraints.
Presented by: Bum Jun Kim
Abstract: Search, ranking, recommendation, knowledge graphs, and retrieval-augmented generation now depend on deep neural networks; reliability under noise, distribution shift, and adversarial manipulation is central to trustworthy deployment. This half-day lecture-primary tutorial presents robustness as a continuous workflow: diagnose expected shifts, stress-test and evaluate honestly, understand failures, design interventions at the data, model, and architecture levels, and report and deploy responsibly. It is a balanced, method-agnostic survey for a general technical CIKM audience, foregrounding safety, search and recommendation integrity, fairness, and drift. Each stage connects to dense retrieval, recommendation under poisoning, knowledge graph and GNN robustness, and RAG. A light, browser-based notebook with an offline fallback makes the checklist concrete using a small dense retriever. Attendees leave with a failure-mode mental model, an evaluation toolkit, and a deployment checklist. No robustness background is required.
Presented by: Konstantinos Skitsas, Davide Mottin and Panagiotis Karras
Abstract: Identifying correspondences between nodes across graphs is a fundamental problem in data management, network analysis, and machine learning, with applications ranging from cross-platform social network de-anonymization to drug discovery and financial fraud detection. This half-day tutorial offers a unified, comprehensive analysis of two complementary graph correspondence paradigms: graph alignment and subgraph matching. For graph alignment, we systematically cover the major algorithmic families, including QAP-based, spectral, embedding-based, and optimal transport methods, across unrestricted, attributed, and seeded settings. For subgraph matching, we follow the canonical filter, order, enumerate pipeline, detailing state-of-the-art optimizations in each stage. Throughout the tutorial, we emphasize underlying problem formulations, algorithmic trade-offs, and practical guidance for selecting the right method for a given task.
Presented by: Fabian Hinder, Valerie Vaquet and Philip Naumann
Abstract: Monitoring systems analyze live data to detect and characterize anomalies and failures, enabling informed decision-making. As systems become increasingly complex, the ability to capture interactions among multiple features in real time is growing in importance across domains. In this tutorial, we approach monitoring through the lens of data streams, concept drift, and explainability. We begin with methods for automatically detecting anomalous behavior and then turn to machine-assisted analysis techniques that provide insight into observed changes, including their geometry and potential root causes. We cover both practical algorithms and theoretical foundations, and place particular emphasis on two complementary perspectives on data shift explainability: (1) the geometry of change, which introduces optimal transport as a model of data shift and leverages Wasserstein distances to characterize distributional changes; and (2) the causality of change, which extends computational causal methods for actionable root-cause analysis. Finally, we demonstrate an end-to-end workflow for critical infrastructure applications.
Presented by: Roger Zhe Li, Aghiles Salah, Mete Sertkan, Changhong Jin, Yingjie Niu, Ruihai Dong and Barry Smyth
Abstract: Traditional recommender systems rely on arbitrary integer IDs to identify items, which carry no semantic information and force models to learn representations from interaction data. While leveraging content information can partially mitigate cold-start, this paradigm inherently suffers from exploding embedding tables and scenario-specific architectures. Recent years have seen the emergence of semantic IDs, which encode semantic information directly into the item identity, reflecting content, hierarchy, and relationships. This framework addresses the above limitations by offering a principled way to encode large-scale semantic information into item identity, benefiting both generative and discriminative applications and unlocking more native unification across modalities, domains, and tasks. This tutorial traces the evolution from raw IDs to semantic IDs as a scalable means of information utilization, offering a comprehensive review of their development and applications across multi-modality, cold-start, and cross-domain recommendation. Hands-on exercises guide attendees through building and evaluating semantic IDs, training models, and exploring cross-domain transfer. We conclude by discussing open challenges, including dynamic semantic IDs and multi-stakeholder utility.
Presented by: Shao Tang, Aman Gupta, Kevin Rossell, Edesio Alcobaca, Lars Hertel, Imran Razzak, Flora Salim, Luis Moneda, Herbert Fei, Daniel Silva and Rohan Ramanath
Abstract: Large language model (LLM)-based agents are moving rapidly from demos to production, but the methodology for getting them there reliably remains fragmented: architecture, observability, evaluation, and online measurement are usually taught and built in isolation. This half-day tutorial presents a practical playbook for building, evaluating, and deploying LLM agents in high-stakes production settings like healthcare and finance. We cover modular context engineering, user simulation, tool and knowledge design, and – as the centerpiece – rigorous evaluation and harness optimization: calibrated LLM-as-a-judge pipelines, inter-rater reliability, and automated judge-prompt optimization, culminating in offline-online correlation that lets teams predict production impact before shipping. Principles are grounded in two regulated, high-stakes domains: customer support in finance and agentic AI in healthcare. Attendees leave with reusable recipes, evaluation code patterns, and a clear mental model connecting offline development to online outcomes. No production experience is required.
Presented by: Ludovico Boratto, Gianni Fenu, Francesca Maridina Malloci, Mirko Marras, Giacomo Medda and Alessandro Soccol
Abstract: Recommender systems are central to information access, discovery, and decision support in modern knowledge-intensive systems. Beyond predicting user–item interactions, they increasingly need to reason over heterogeneous entities, relations, and contextual evidence while remaining transparent, trustworthy, and reusable. Path reasoning over knowledge graphs provides a principled foundation for this goal, as it connects users, items, and knowledge through semantically meaningful chains that can support explainable recommendation. However, adoption remains limited by fragmented implementations, heterogeneous preprocessing pipelines, and high coding complexity. This tutorial introduces hopwise, a framework for building explainable-by-design recommender systems through path reasoning on knowledge graphs. Participants will follow a progressive hands-on workflow: transforming recommendation datasets into knowledge graphs, running reinforcement learning–based and language model–based reasoning methods, evaluating their utility and explainability, and extending the framework.
Presented by: Chuan Meng, Litu Ou, Sean MacAvaney and Jeff Dalton
Abstract: The information retrieval (IR) community is moving from traditional human-driven search towards agentic search, where agents iteratively generate search queries, inspect retrieved documents, and repeat the process until sufficient evidence is gathered to produce a final answer. Agentic search is important because it can scale knowledge-intensive work, free up valuable time for humans, and enable agents to interact with external knowledge. To the best of our knowledge, this is the first dedicated tutorial on agentic search. The tutorial will cover foundations and systematic optimization paradigms, as well as advanced topics including long-horizon agentic search (deep research). It is designed for students, researchers, and practitioners from both academia and industry. Participants will gain a comprehensive understanding of agentic search, together with a live demonstration and hands-on experience in running a practical agentic search system.
Presented by: Amin Bigdeli, Radin Hamidi Rad, Hai Son Le, Yibin Lei, Andrew Yates, Negar Arabzadeh, Charles L. A. Clarke and Ebrahim Bagheri
Abstract: The emergence of large language models and agentic search systems is transforming query expansion, elevating it from a standalone preprocessing step to a central reasoning component in modern retrieval pipelines. The shift from classical term-selection methods to LLM-based generation has opened new possibilities for expansion, while deep search agents, retrieval-augmented generation, and conversational retrieval have introduced fundamentally new settings in which query expansion must operate. This tutorial offers a comprehensive treatment of query expansion methods, tracing their progression from classical statistical techniques through neural and transformer-based approaches to LLM-based generation, while covering evaluation methodology, cross-paradigm deployment, failure analysis, and emerging research opportunities. The tutorial is accompanied by hands-on experimentation for implementing and comparing expansion methods across retrieval settings.
Presented by: Andrew Yates, Carlos Lassance, Cosimo Rulli, Eugene Yang, Sean MacAvaney, Siddharth Singh, Thong Nguyen and Yibin Lei
Abstract: Learned Sparse Retrieval (LSR) techniques use neural machinery to represent queries and documents as learned bags of words. LSR has been shown to be a robust and efficient family of methods for retrieving high-quality search results. This half-day tutorial aims to provide an extensive overview of LSR, ranging from its fundamentals to the latest emerging techniques. By the end of the tutorial, attendees will be familiar with the important design decisions of an LSR model, know how to apply them to text and other modalities, and understand the latest techniques for retrieving with them efficiently. Website: https://lsr-tutorial.