Building a Large Music Recommender Leveraging AI, Deep Learning and Human Expertise

Explore the intricate world of large-scale music recommendation systems in this 34-minute conference talk by Òscar Celma from Pandora. Delve into the evolution of The Music Genome Project and how it combines with vast user data to create personalized radio stations. Learn about the interdisciplinary approach Pandora employs to analyze massive datasets, balancing familiarity, discovery, and relevance for individual listeners. Gain insights into the application of Machine Learning (ML) techniques for music recommendation, including online and offline ML architecture. Understand how user satisfaction is measured and evaluated, and discover the critical factors in selecting the right song for the right listener at the right time. The talk covers lessons learned, big data challenges, model selection, business goals, evaluation methods, experimentation, and the importance of domain expertise in music recommendation systems. Conclude with a summary and Q&A session to deepen your understanding of AI-driven music recommendation at scale.

Introduction
Welcome
What is Pandora
Lessons Learned
Big Data
Model Selection
Business Goals
Offline Evaluation
Online Evaluation
Experimentation and Production
Framework
Domain Expertise
Why is Music Different
Dont do stupid things
Summary
QA