Being in the price comparison business, solute's mission is to make sense of product data. Crucial to fulfilling that mission is figuring out the category of each offer. We tackle this problem with Machine Learning algorithms for over 10(!) years now.
We want to invite you to a journey through our history of building, maintaining, and modernizing our category classification system. Starting from back in the days where people used blackberry phones and scikit-learn wasn't even invented yet. You will learn about the rise of our SVM classifier, well-founded decisions leading to a successful system that just needed some tweaking over the years — until this approach didn't suffice any more. We will share our most interesting mistakes, misconceptions and design flaws and how we moved forward with our rework of the solute Machine Learning infrastructure and the introduction of a Neural Network based category classifier. No previous knowledge of Machine Learning algorithms is required.
I studied Natural Language Processing and Computer Science, worked some years at a university. After that I ended up being a software developer in Karlsruhe at solute GmbH where I can play around with millions of product data every day and try to make sense of it. I like podcasts, board games and word plays.