How to Solve Product Classification Challenge for Marketplaces in the Wild
In the pursuit of improved user experience, marketplaces are trying to solve open issues using modern approaches. Today, it seems that AI can easily solve a lot of narrow problems but when it faces real-world questions, it turns out that those questions are much deeper than they seem at first glance. Since we are following current industry dynamics in AI, we have started with image classification. We’ll tell you about our experience with real-world problems associated with this task. How we went from a flat-model to one that is ‘super-complex’ as well as what was successful and what wasn’t.
Denis Troyanov is a young researcher at Lalafo, a mobile C2C marketplace powered by artificial intelligence. After earning a degree in computer science at South-Ural State University, Denis moved to Kyiv, Ukraine to work as a research engineer with an emphasis on deep learning. He has completed several projects in computer vision and deep learning and worked with classical computer vision for a 'face' and gesture processing company with DL models for the fashion and entertainment industry. Denis is currently working on very deep classification for 10k+ classes for e-commerce under product constraints such as unbalanced classes, weak labelling, constantly evolving catalog of classes etc. Denis is passionate about continuing research in machine learning and topics such as psychology, game theory, reinforcement learning, how memory works and decision making concepts.