Learning representations of customer behavior to drive actionable insights in e-commerce

Rubikloud enables online businesses to turn their data into revenue by turning the focus from visualization and interpretation to driving smarter data-driven decisions. A key challenge to leveraging the advances in machine learning research and development is the nature of event-based data encountered in this domain: clicks, purchases, impressions, and conversions. Machine learning techniques typically operate on fixed-length vector representations of data, for example, collections of attributes, images, and word counts of text documents. They are not designed to cope with rich histories of behaviours. This research project proposes to develop representation learning methods that can process histories of events, so-called “irregular data”. We will learn abstractions from event data that will enable prediction, trending, attribution, and other actionable insights.

Intern: 
Hojjat Salehinejad
Superviseur universitaire: 
Dr. Graham Taylor
Project Year: 
2014
Province: 
Ontario
Université: 
Discipline: 
Programme: