|Home||Keynote||Program||Submission Instructions||Program Committee|
Speaker: Aapo Kyrola, CMU
Title: Large-scale recommender systems on just a PC
GraphChi [Kyrola et. al., OSDI 2012] is a disk-based system for large-scale graph computation that is able to handle graphs with billions of nodes and edges, on just a laptop or PC. GraphChi has become a popular platform for research in recommender algorithms thanks to the comprehensive collaborative filtering toolkit developed by Danny Bickson, and its ability to solve very large problems with limited amount of RAM – thus saving practicioners hurdles of using distributed computation. To be published in RecSys this year, GraphChi will also add support for large-scale random walk computations – a staple of many recommendation algorithms for social networks and other natural graphs.
In this keynote I will give a short introduction to GraphChi and its collaborative filtering toolkit. I will then go into detail on how to recommender algorithms on top of GraphChi are implemented. For practitioners, evaluation and testing of different techniques is still too manual and error-prone task, so I will end by proposing directions to further unify different implementations of recommender algorithms to facilitate ad-hoc evaluation of different techniques and/or parameter settings automatically.
Speaker: Justin Basilico, Lead Researcher, Netflix
Justin Basilico is a Lead Researcher/Engineer in the Personalization Science and Engineering group at Netflix where he does applied research at the intersection of machine learning, ranking, recommendation, and large-scale software engineering. Prior to Netflix, he worked on machine learning in the Cognitive Systems group at Sandia National Laboratories. He is also the co-creator of the Cognitive Foundry, an open-source software library for building machine learning algorithms and applications.
Title: Recommendation at Netflix Scale
Netflix instant video streaming represents an estimated one third of peak broadband traffic in the US. Personalization is at the core of our product with recommendations driving about 75% of all viewing. Building a high-quality recommendation system for millions of users requires a careful balancing act of handling large volumes of data, keeping recommendations fresh and accurate, remaining responsive to user actions, and also being flexible to accommodate research and experimentation. In this talk, I will discuss our approach to recommendations at Netflix and our cloud-based architecture for splitting up computation among offline, nearline, and online regimes. This allows us to balance accuracy, freshness, and responsiveness of the recommendations.