Learning customers' preferences in (almost) real time
27 maggio 2019 alle 9:00
Presso l'Aula 208, Settore Didattico, via Celoria 20
Speaker: Giovanni Zappella, Amazon Music, Berlin (DE)
Presona di riferimento: Nicolò Cesa-Bianchi.
Personalization is a crucial aspect of many online experiences. In particular, content ranking is often a key component in delivering sophisticated personalization results. In this talk, I will use a real personalization use-case from the music domain to explain how it is possible to learn customers' preference in (almost) real time and which kind of effort it requires. I will provide some insights in the reasoning behind the choices that were made for this specific use-case and explain how they translate in a real production system. I will also try to summarize some general concepts and best practices which can help to drive the reasoning about machine learning solutions when employed in “the real world”.
Giovanni Zappella is a Sr. Applied Scientist at Amazon Music. After obtaining his PhD at the Università degli Studi di Milano, he covered different positions including Machine Learning Scientist in the Amazon Core Machine Learning team and team lead of the Artificial Intelligence team at UniCredit. His current work focuses on designing, testing and deploying solutions trading off exploration and exploitation in order to provide personalised content in user-facing applications and accelerating online experimentation.