Products-Classification

I solved the test task from KazanExpress on multiclass products classification. Each product has its title, description, characteristics, images and other data that is usually included into product cards on the marketplace. I developed a multimodal model which uses embeddings from ruBERT, FastText, Wiki2Vec and feature maps obtained from ResNet pretrained on the ImageNet. My solution performs 0.884 on f1 weighted score on more than 750 classes.