Spotify - Personalising the Listening Experience - Mounia Lalmas

Explore the intricacies of personalizing music recommendations in this conference talk from the Alan Turing Institute. Delve into Spotify's approach to matching fans with artists in a relevant and personalized manner. Learn about datasets, shared tasks, and the implementation of machine learning for Spotify Home. Discover the BaRT (Bandits for Recommendations as Treatments) algorithm and its application in multi-armed bandit scenarios. Understand how success is measured through reward functions and how these are personalized based on user preferences. Examine experiments conducted to improve personalization, including the analysis of user intents on Spotify Home and modeling techniques. Investigate the delicate balance between relevance, diversity, and user satisfaction in music recommendations. Gain insights into the ongoing efforts to enhance the personalized listening experience for Spotify users.

Intro
What does it mean to match fans and artists in a personal and relevant way?
Datasets and shared tasks
Making machine learning work for Spotify Home
BaRT (Bandits for Recommendations as Treatments)
BaRT: Multi-armed bandit algorithm for Home
Success is captured by the reward function
Personalizing the reward function
Most frequent genre in co-clusters
Reward function (success) per co-cluster
Experiments
Towards personalizing with respect to success
User intents on Home
Modeling user intents for Home
Towards personalizing with respect to user intents
The interplay between relevance, diversity & satisfaction
Towards personalizing with respect to diversity
Personalizing the listening experience