E-Commerce Product Categorization

In the modern E-Commerce environmental  product listings on web sites can be ever updating. And to drive customers to the products that they are ready to spend money in, a website must be able to quickly categorize and cross list an item.  In the paper "Large Scale Itemization in E-commerce" the authors use the example of EBay in demonstrating the need to be able to categorize large numbers of items for easy understanding.

EBay lives off the fees collected when a auctions are placed on the site. Its success depends on their ability to drive paying customers to the listed on their site. Take for instance a male customer wants to buy dress socks. He could search for socks and find  socks for men and women, dress or casual, and  many other categories. EBay needs to create categories for products, but does not want the categories to be to specific or to general.

The paper details a process that creates a three stage categorization process for clustering groups of similar items. The first step is to create sets of mutually exclusive "latent groups" that all the products could be broken up into. Then a course level categorization places all products into one of the latent groups. Finally, within the groups a second fine level categorization is conducted to assign all products in that latent group in to a class.

Using the socks example started above, EBay would first categorize all the products for sale on the site. All socks would be placed into a latent group for socks. Then all the socks would be classified further into the different types of socks.

The paper found that when this "novel" method was compared to more conventional methods of hierarchical clustering, it would improve classification while maintaining the speed of classification. This can have significant advantages to an e-commerce site that handles large amounts of product listings that last only a short period of time.

Source: http://labs.ebay.com/wp-content/uploads/2012/10/Jean-David-Ruvini-Large-scale_Item.pdf