Matrix Factorisation for Recommender Systems

Venue Capacity: 70 people The capacity of the venue is about 70 people but because of no-shows, we have a policy of first-come, first-served on the night in the event of the room filling up. Managing no-shows is a problem all Meetups have to deal with and this is the fairest method we can think of. While we have never been in a position to turn anyone away for capacity reasons, it is always a possibility, so please arrive early to avoid disappointment on the night. Eric Mehes is a Data Scientist/Research Engineer at Zalando Dublin. His area of work is split between building recommender systems to streamline the experience of Zalando users and creating audiences that would maximise the performance of various Zalando marketing campaigns. This talk will give a brief introduction to the area of collaborative filtering techniques used in recommender systems. The often utilised Alternating Least Squares, or ALS (Matrix Factorization Techniques for Recommender Systems, Yehuda Koren et al, 2009) algorithm will be presented in depth. This technique was used (along with restricted boltzmann machines) in the algorithm that ended up winning the Netflix Prize. A few possible metrics used to gauge the performance of a recommender system will also be shown. The aim of the talk is to give the audience a clear understanding on how ALS works, what are it’s possible gotchas, and what avenues can one take in optimising an ALS-based recommender system.