Tuesday, November 8th, 2011
How can you rank product search results when you have very little data about how past shoppers have interacted with the products? Through large scale analysis of its clickstream data, Etsy is automatically discovering product attributes (things like materials, prices, or text features) which signal that a search result is particularly relevant (or irrelevant) to a given query. This attribute-level approach makes it possible to appropriately rank products in search results- even if those products are brand new and one-of-a-kind. This presentation discusses Etsy’s efforts to predict relevance in product search, in which Hadoop is a central component.