Learning to Model Sequences:
Playlist Prediction for Local Music Discovery

Project Goals

The ability to learn predictive models of sequences is a component in several application problems, ranging from language models for machine translation to the recommendation of playlists in online media systems. The goal of this project is to develop new machine learning algorithms that can learn sequence models for items that are difficult to describe by attributes. In particular, the project develops models that automatically embed items in a latent feature space based on training sequences, that can integrate partial and noisy side information, and that have the ability to model long-range dependencies. While the resulting models span a wide range of applications, the project focuses on the recommendation of music playlists as the main testbed. In particular, the project will develop and deploy an online music recommendation system for local music discovery.

People

• Thorsten Joachims (PI)
• Douglas Turnbull (PI)
• Shuo Chen (Ph.D. Student)
• Joshua Moore (Ph.D. Student)
• Tobias Schnabel (Ph.D. Student)
• Jisha Kambo (Undergraduate Student)
• Nohemi Fernandez (Undergraduate Student)
• Ziyu Fan (Undergraduate Student)
• Warren Crowell (Undergraduate Student)
• Akilesh Potti (Undergraduate Student)
• Justin Zupnick (Visiting Master Student)
• Navin Chandak (Visiting Summer Intern)
• Shane Moore (Undergraduate Student)
• Sid Reddy (Undergraduate Student)
• Ashudeep Singh (Visiting Summer Intern)
• Matthew Kercher (Undergraduate Student)
• Mariah Flaim (Undergraduate Student)
• David Dorsey (Undergraduate Student)
• Dan Kustin (Undergraduate Student)
• Adam Linden (Undergraduate Student)
• Chris Perez (Undergraduate Student)
• Luke Waldner (Undergraduate Student)
• Ali Akhtar (Undergraduate Student)
• Andrew Horwitz (Undergraduate Student)
• Stephen Meyerhofer (Undergraduate Student)
• Alex Spirgel (Undergraduate Student)
• Kris Stensland (Undergraduate Student)
• Laurence Welch (Undergraduate Student)
• Alex Wolf (Undergraduate Student)
• Jon Burger (Undergraduate Student)

Demos and Systems

• MegsRadio Local Music Discovery System
• LME playlist Generation Demo

Software and Data

• LME Sequence Embedding Software
• LME Sequence Embedding Software using Portal-Model for Parallelization
• Yes.com Playlist Dataset
• Blade/Chest Embedding Software for Modeling Intransitive Preference Relations

Publications

Related Courses

• Search Engines and Recommender Systems, Spring 2015. (course material)

• Advanced Machine Learning, Spring 2015. (course material)

• Machine Learning, Fall 2014. (course material)

• Advanced Topics in Machine Learning, Spring 2014. (course material)

Related Presentations

• T. Joachims. Learning from User Interactions through Interventions. Invited Keynote Talk at the ACM International Conference on Web Search and Data Mining (WSDM), 2015.
• Shuo Chen, T. Joachims. Modeling Intransitivity in Matchup and Comparison Data. Invited talk at the KDD 2015 Workshop on Large-Scale Sports Analytics, 2015.
• T. Joachims. Online and Batch Learning with Interventions. Invited Keynote Talk at the Pacific-Asia Conference on Knowledge Discovery and Data Mining (PAKDD), 2015.
• S. Reddy. Learning Representations of Student Knowledge and Educational Content. Invited talk at the ICML Workshop on Machine Learning for Education, 2015.
• D. Turnbull. Growing Local Music. Invited talk at TEDx Ithaca College, 2014.
• T. Joachims. Learning from Rational Behavior. Invited Keynote Talk at the Conference on Empirical Methods in Natural Language Processing (EMNLP), 2014.
• D. Turnbull. Google, Netflix, Pandora: How to Build Your Own Recommender System. Science talks at Ithaca High School (2014) and Dryden High School, 2013.
• D. Turnbull. Constructive Music Information Retrieval. Invited talk that the NIPS Workshop on Constructive Machine Learning, 2013.

Acknowledgement and Disclaimer

This material is based upon work supported by the National Science Foundation under Awards IIS-1217686 / IIS-1217485. Any opinions, findings and conclusions or recommendations expressed in this material are those of the author and do not necessarily reflect the views of the National Science Foundation (NSF).

Last change: 1/29/2017