Market: How Are Products Curated Into Categories?

We use a data science model based on machine learning, which has learned keywords relating to each of our unique categories. It can scan the product descriptions and use this information to place each product into the most relevant category by the terminology found.

Curation Methodology

EDITED collects products from global retail markets across apparel, homeware and beauty. These products are sold on e-commerce platforms that employ an incredibly diverse set of practices. They are presented in their own language and local currency. Some are organised mirroring internal company structures (e.g. dress vs tops), some by material (e.g. knit vs jersey), and others might still be split by purpose, (e.g. work vs active). An outlet store, a specialty retailer, and a department store may categorize and describe the exact same product in different ways in their corresponding e-commerce stores.

To make sense of our diverse and inherently noisy data set, EDITED Market uses its own comprehensive classification of products, made up of categories, subcategories, styles and details. We also normalise and augment certain product characteristics such as colour codes, patterns and some sizes. This allows users to work with a normalised data set helping to compare like for like.

There may be instances when a product within Market sits in the wrong classification group, and as a user you may spot this ahead of our QA team. To assist you with this EDITED allows for you to amend curations within the app.

To do this, simply click the 'X' next to the incorrect category and choose the one you think the product best fits into.

Amending curations to be correct supports our data quality and improves the algorithms we use by strengthening their knowledge.

When you amend a curation we run it through our data pipe. If data pipe is working on a new classification we are introducing it may already have products waiting, when this is the case your curation will join a queue and could take up to 72 hours to show correctly. When traffic is less busy it may show quicker than this.