Very excited to have won first place on Dotz’ s first Data Science Competition!
Alongside my great teammate Tobias Veiga, we developed the best solution for hierarchical classification of e-commerce products using their textual descriptions.
Always interesting to develop a NLP solution in a language different than English and bridge the gap between state of the art and industry - Our solution was praised not only for the highest metrics achieved but for it’s scalability and ease of use. Our pipeline used a mixture of portuguese Bert, Character CNNs, traditional sparseBOW model and hierarchical specific models.
Shout out to fellow contenders Felipe Maia Polo and André Sacilotti for their amazing solutions as well, it was a very close competition and I’ll be keeping a eye on what they develop in the future. More info (portuguese)