Abstract: Recent decades have witnessed a great proliferation of recommendation systems, which have found application in many business verticals. It is however challenging for practitioners to select and customize the optimal algorithms for a specific business scenario. In addition to training and tuning the appropriate algorithms, a complete recommendation system also consists of operations such as data pre-processing, model training, model evaluation and system operationalization.
Motivated by our extensive experience in productization of recommendation systems in a variety of real-world application domains, in this talk, we will review complete pipelines of building recommendation systems. We will start by introducing some standard factorization machine algorithms. Thereafter, we will address some of the latest advances in deep learning algorithms in the area, with an emphasis on knowledge graph models. Then we will analyse different methodologies for computing these algorithms at scale, reviewing some available techniques for hyperparameter tuning. Finally, we will discuss how these systems can be brought successfully into production.
To support this talk, an extensive suite of algorithms, utilities and Jupyter notebooks are open source and publicly available in our Recommenders repository (https://github.com/Microsoft/Recommenders).
Bio: Andreas Argyriou is a Senior Data Scientist with the Azure Customer Advisory Team at Microsoft. Before that, he was a Senior Data Scientist at Kayak.com and held various positions in academic research. He has published work on multitask and kernel-based learning, sparse regularization and convex optimization in top conferences and journals in machine learning. He obtained a PhD in machine learning from University College London and a BSc and MEng in computer science from MIT. His current work focuses on algorithms for production-ready systems and applications at scale such as recommendations systems.