RecSys 2025 edition. This is the original industrial tutorial presented at RecSys 2025. See also the SIGIR 2026 edition.
About the tutorial
Modern recommender systems have made major progress, but many large-scale, user-facing solutions still behave like static “one-shot” recommenders. As user expectations shift toward interactive, context-aware, and adaptive experiences, recommender systems increasingly need to support multi-step reasoning, tool use, memory, and autonomous orchestration.
This tutorial focuses on how recent advances in large language models (LLMs) enable a new class of recommender systems: agentic (and often multi-agent) systems that can:
- reason over evolving user needs,
- interact through dialog and longer context,
- call tools and external APIs,
- orchestrate multi-step workflows,
- refine outputs using constraints and feedback,
- and support practical production requirements (reliability, scalability, transparency, safety).
Examples and patterns discussed include context-aware recommendation, dynamic multi-step orchestration, and personalized recommendation pipelines, culminating in a hands-on session that bridges concepts with implementation.
Instructors
| Instructor | Affiliation |
|---|---|
| Reza Yousefi Maragheh | Walmart Global Tech |
| Yashar Deldjoo | Polytechnic University of Bari |
| Chi Wang | Google DeepMind |
| Jason Cho | Walmart Global Tech |
| Derek Cheng | Google DeepMind |
What you’ll learn
By the end of this tutorial, you should be able to:
- Understand the shift from traditional recommenders to LLM-powered, interactive, agentic systems
- Recognize the core building blocks (“alphabets”) of multi-agentic systems, including:
- memory types and retrieval strategies
- function calling and tool usage
- orchestration protocols / interfaces
- reasoning load balancing across steps or agents
- Apply common agentic RecSys design patterns for:
- conversational recommendation
- context-aware autonomous recommendation
- recommendation evaluation and user simulation
- explanation generation
- Gain hands-on familiarity with frameworks commonly used for agentic pipelines (e.g., multi-agent and orchestration frameworks)
- Identify practical pitfalls and open challenges such as:
- scalability and latency constraints
- hallucinations and error propagation
- transparency, fairness, bias, and privacy risks
Who this is for
This tutorial is designed for:
- PhD students and researchers exploring agentic systems for recommendation
- Senior researchers and practitioners working with generative/LLM-based RecSys
- Industry teams looking for practical patterns to move from prototypes to scalable systems
Materials and companion resources
- Slides (PDF) — full tutorial deck
- architecture diagrams
Citation
If you find this tutorial useful in your research or work, feel free to cite our tutorial:
@inproceedings{yousefi2025multi,
title={Multi-Agentic Recommender Systems: Foundations, Design Patterns, and E-Commerce Applications—An Industrial Tutorial},
author={Yousefi Maragheh, Reza and Deldjoo, Yashar and Wang, Chi and Cho, Jason and Cheng, Derek},
booktitle={Proceedings of the Nineteenth ACM Conference on Recommender Systems},
pages={1427--1429},
year={2025}
}
Yousefi Maragheh, R., Deldjoo, Y., Wang, C., Cho, J., & Cheng, D. (2025). Multi-Agentic Recommender Systems: Foundations, Design Patterns, and E-Commerce Applications—An Industrial Tutorial. In Proceedings of the Nineteenth ACM Conference on Recommender Systems (pp. 1427–1429).