Description
During this talk, I will share my experience gathered during the development of Allegro's visual search engine. I will present our entire journey from the plain idea through different modelling approaches up to its current solution. Finally, I will provide some tips and tricks that we have learned along the way and show a lot of images - after all, that's what you're looking for!
How can one effectively find a recently seen t-shirt on an e-commerce platform such as Allegro? Or maybe an unusual set of cups, that is used by the coffee shop downstairs? One could try to describe them, but a picture is worth a thousand words - what if that one picture or actually a photo was enough? With over 100 million offers and with multiple photos for every product, this seems nearly impossible. Especially given how similar some products are. During this talk, I will share how we make it possible by introducing a visual search model. I will focus on our machine learning model and its evolution over time but also cover some technical aspects of our approach.