I am searching for an open-source (or cheap/easily available) dataset of consumer package goods that contains information about brand/item attributes, e.g., size, color, flavor, package type. Ideally I'd like an acyclic directed graph showing attributes at varying levels of abstraction -- think subclasses in OO terms, e.g., "Tide Pods Ocean Mist 77ct" inherits attributes of it's parent "Tide" which in turn inherits from it's parent "Laundry Detergent".
Specifically I'd like to leverage product attribute information to deal with the sparsity problem in product recommendation systems as discussed in the section 'Content Description' of this article on collaborative filtering. I'd prefer to do this using an existing dataset rather than creating a bag-of-words attribute set by parsing product descriptions.
This closed ticket points to the product open data repository which is an extensive dataset, but whose fields are limited to size, weight, and volume. And there is the Factual database, but it appears limited to food stuffs and nutritional information.