Laurent Charlin / MILA

Conversational systems provide an expressive interface for interacting with users. Combined with a recommender system they enable elicitation of short- and long-term user preferences enabling more precise user modelling. Further they can provide explanations and provide a useful environment for learning multi-step recommendation policies. In this talk, we introduce conversational recommender systems, survey current models and datasets, and discuss open problems in light of the availability of large-scale pre-trained language models.
Speaker Bio: Laurent Charlin is a Canada CIFAR AI Chair at Mila and an associate professor at HEC, the business school at the University of Montreal. His research focuses on developing novel machine learning models to aid in decision making. His recent work has focussed on continual learning and on applications in fields such as recommender systems and optimization. He has a number of highly cited publications in dialogue systems. Laurent co-developed the Toronto Paper Matching System (TPMS) with Richard Zemel.
Jaya Kawale / Tubi

In this talk, I will discuss the journey of creating a personalized binge-watching recommendation system based on real-time multi-interest based retrieval at Tubi. I will explain the challenges faced in creating such a large scale system, including the need to understand users' multiple interests and preferences, as well as the need for real-time processing to ensure that recommendations are delivered quickly and accurately. I will also discuss the generating of various embeddings using multiple machine learning and natural language processing algorithms and techniques.
Speaker Bio: Vice President Of Engineering (Machine Learning) at Tubi
Eva Zangerle / University of Innsbruck, Austria

Music streaming platforms provide access to millions of songs. They have significantly impacted not only the way people listen to music but also how users discover and explore new music. In this talk, I will review what we can (and cannot) learn from the analysis of music streaming data. I will provide several examples of studies that leveraged music streaming data to get a deeper understanding of users, their behavior, preferences, the effects of personalization, trends in music consumption, or the performance and limitations of current personalization algorithms. Furthermore, I will also critically reflect on the limitations of using data gathered from music streaming platforms for such analyses.
Speaker Bio: Eva is an assistant professor at the Department of Computer Science at the University of Innsbruck, Austria. Her primary scientific interests focus on (context-aware) recommender systems and user modeling aspects of music information retrieval tasks. She earned her Ph.D. from the University of Innsbruck in the field of recommender systems for collaborative social media platforms. During her postdoc, she did short-term research stays at Ritsumeikan University in Kyoto, Japan (funded by a Postdoctoral Fellowship for Overseas Researchers from the Japan Society for the Promotion of Science), Freie Universität Berlin, Germany (funded by the Global Faculty Program of Freie Universität) and Johannes-Kepler-Universität Linz, Austria.
Ben Carterette / Spotify

One of Spotify’s missions is “to match fans and creators in a personal and relevant way”. This talk will share some of the research work aimed at achieving this, from using machine learning to metric validation, and illustrated through examples within the context of Spotify’s home and search. An important aspect will focus on illustrating that when aiming to personalize for both recommendation and search, it is important to consider the heterogeneity of both listener and content. One way to do this is to consider the following three angles when developing machine learning solutions for personalization: (1) Understanding user journey; (2) Optimizing for the right metric; and (3) Thinking about diversity.
Speaker Bio: Ben Carterette is a Sr Research Manager at Spotify, where he leads teams of research scientists in their work on music & podcast search, recommendation, representation, and evaluation. Formerly an Associate Professor at the University of Delaware, Ben has published over 150 papers in venues such as SIGIR, WSDM, KDD, CIKM, and the Web Conference and won several Best Paper Awards. He served as Chair of the ACM SIGIR Executive Committee from 2019 to 2022.

Speaker Bio: None