We will share our experience gained during development of ‘learning to rank’ system in an e-commerce setting. Our talk will cover both theoretical and practical aspect of the project - general approach in learning to rank, suitable tools, data and modelling pipeline and production deployment considerations. Finally, we will share some tips and tricks that come from actual, practical insights.
With almost 100 million offers coming from a diversified set of categories and 40 million searches daily - it is a challenge to address the need of our customers - identify the most relevant offer and present it on the top of the results list. Being able to do it with both high quality and low latency makes it even harder. Apart from looking at the technical side of developing a ‘learning to rank’ system - we will additionally show the evolution of our experimentation and modelling processes.