Product ontologies - consisting of a taxonomic categorization of product types and lists of attributes that types and products have - are invaluable for analyzing sales, opinions and ratings of items on e-commerce sites. Unfortunately, many international and smaller sites lack such ontologies, and instead feature only coarse high-level categories. We present a Siamese neural model which utilizes such coarse categories to learn a fine-grained hierarchical categorization of products, and jointly extract lists of product attributes from text reviews. We show that our model retains a high accuracy on the categorization task for unseen products and unseen category depths, and as a side effect learns to extract useful product attributes.