RecSys 2021 Tutorial on Conversational Recommendation: Formulation, Methods, and Evaluation Wenqiang Lei National University of Singapore Singapore wenqianglei@gmail.com Chongming Gao University of Science and Technology of China China chongming.gao@gmail.com Maarten de Rijke University of Amsterdam & Ahold Delhaize The Netherlands m.derijke@uva.nl ABSTRACT Recommender systems have demonstrated great success in infor- mation seeking. However, traditional recommender systems work in a static way, estimating user preferences on items from past inter- action history. This prevents recommender systems from capturing dynamic and fne-grained preferences of users. Conversational rec- ommender systems bring a revolution to existing recommender systems. They are able to communicate with users through natural language, which enables them to explicitly elicit user preferences by asking whether a user likes an attribute or item or not. Based on in- formation shared through users’ responses, a recommender system can produce more accurate and personalized recommendations. We identify fve emerging trends in the general area of conver- sational recommender systems: (1) Question-based user preference elicitation; (2) Multi-turn conversational recommendation strate- gies; (3) Dialogue understanding and generation; (4) Exploitation- exploration trade-ofs; and (5) Evaluation and user simulation. This tutorial covers these fve directions, providing a review of existing approaches and progress on each topic. By presenting the emerging and promising topic of conversa- tional recommender systems, we aim to provide take-aways to practitioners to build their own systems. We also want to stimulate more ideas and discussions with audiences on core problems of this topic such as task formalization, dataset collection, algorithm development, and evaluation, with the ambition of facilitating the development of conversational recommender systems. ACM Reference Format: Wenqiang Lei, Chongming Gao, and Maarten de Rijke. 2021. RecSys 2021 Tutorial on Conversational Recommendation: Formulation, Methods, and Evaluation. In Fifthteenth ACM Conference on Recommender Systems (RecSys ’21), September 27-October 1, 2021, Amsterdam, Netherlands. ACM, New York, NY, USA, 3 pages. https://doi.org/10.1145/3460231.3473325 1 INTRODUCTION Recommender systems have become the prevalent choice for infor- mation seeking. Efective recommendations that are both accurate Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for proft or commercial advantage and that copies bear this notice and the full citation on the frst page. Copyrights for components of this work owned by others than the author(s) must be honored. Abstracting with credit is permitted. To copy otherwise, or republish, to post on servers or to redistribute to lists, requires prior specifc permission and/or a fee. Request permissions from permissions@acm.org. RecSys ’21, September 27-October 1, 2021, Amsterdam, Netherlands © 2021 Copyright held by the owner/author(s). Publication rights licensed to ACM. ACM ISBN 978-1-4503-8458-2/21/09. . . $15.00 https://doi.org/10.1145/3460231.3473325 and timely can help users fnd the information they desire, greatly save their time, and bring signifcant value to business. Motivation of conversational recommendations. Traditional recommender systems, which we call static recommender systems in this tutorial, primarily predict a user’s preference towards an item by analyzing their past behavior, e.g., click history, visit log, and ratings on items. Despite the wide usage, static recommender systems have a fundamental intrinsic limitation: it is not able to capture a user’s current preferences, which may have shifted away from their historical ones. In addition, it is hard to fnd accurate reasons as to why a user interacts with certain items as there is no channel for static recommendation system to ac- cess such information. The emergence of conversational recom- mender systems (CRSs) changes this situation in profound ways [2ś 4, 11, 12, 14, 15, 21, 22, 30, 33, 37, 38, 41]. With CRSs, recommendation becomes an interactive process through which users interact with the recommender by many ways such as natural language. In this manner, a CRS should be able to obtain or infer the dynamic preferences of users from their real-time feedback, which might consist of their direct or indirect descriptions of their needs or of answers to questions posed by the recommender system in a mixed-initiative setup. Such information can help a CRS to adjust its recommendation decision efectively and, we hope, dramatically, enhance its performance. Necessity of this tutorial. Recently, attracted by the power of CRSs, many researchers have devoted to exploring this topic. These eforts are spread across a broad range of task formulation, in diverse settings and application scenarios. Although there is increasing attention on CRSs, some important questions also arise: (1) What are the diferences between CRSs and other ongoing research topics such as task-oriented dialogue systems [6] and interactive recommender systems [1]? (2) What are the main challenges in CRSs and potential future opportunities? By ofering this tutorial, we want to answer these insightful questions and thus beneft both industry and academia: • We aim to provide industrial participants with a broad pic- ture of the current development of CRSs, with key take ways for building realistic conversational recommendations for their own scenario. • We aim to provide academic participants with a comprehen- sive literature review and insightful discussions. By summa- rizing existing assumptions and explorations on the topic of conversational recommendation, together with a review of recent progresses on recommender systems and dialogue 842