Tuesday, October 26, 2010

Why local search sites need taxonomy

Local search sites are all working on the same challenge - how to present the most relevant list of local businesses for any given search query - meaning keyword and location. The challenge is three fold:

1) Business data that is commonly available only comes with a single yellow pages style category - something like liquor stores, "Furniture Retailer", or Restaurants - Italian. This limits the results that can be returned on keyword searches for things like "Bookshelves", "Chardonnay" or "Lasagna".

2) Because business data is assigned to high level categories, local search sites can only present general business lists and are limited in the filtering and drill-down options that can be presented to a user. Ideally, a local search site should be able to present a list of Italian Restaurants, but also allow the user to drill down to those who have Chicken Parmesian on the menu.

3) Business data is sometimes not accurately categorized to begin with, meaning that the category assigned to a business is not correct. This is a common local search data problem.

Taxonomy is valuable for solving these two issues in two separate ways. First, it is crucial that a local search environment return relevant business results for as many keyword variations as possible. To achieve this, a local search site needs to expand the number of keywords in its database - sometimes placed in synonym tables - that are associated with each business category. Let's return to the example above for "Furniture Retailer". A user who searches for furniture store, sofa, couch, bed, bookshelf, metal furniture, armoire, desks, etc. should all be presented with listings from the "Furniture Retailer" category. Without adding this additional terminology to the database, then a search for any of the above keywords, plus all the other types of furniture that I didn't even list, would result in a failed search - or no results at all. Users will not return to a site that does not return results, so it is crucial to get this issue taken care of.

Once a strong vocabulary is in place and the local search site is able to resolve most user queries to a relevant business category, the next step is to enrich business data by tagging individual businesses with specific terminology from the taxonomy. One approach some sites use is to allow businesses to enter free text keyword tags about the business. However, this raises challenges in normalizing the information entered by users. For example, one user may enter a tag "sofa" and another user may enter a tag "couch". Without a normalized and controlled vocabulary, it will be impossible to know that these are the same concept and thus to present a list of all businesses that sell sofas/couches to a user. Also, most users won't spend the time to think of very many relevant keywords so the set of tags may be pretty limited.

Instead, the optimal solution is to use a controlled set of tags that a business or user can select from to append to a business. If you now that a business is a Furniture Store, you can present the business or user, in a self-registration environment, with the set of all the controlled keywords that they may want to append to the business. It will be much easier for the person registering to select from a prepopulated list of tags that can simply be checked versus entering keywords into a free text box. Also, the tags that are selected with be clean and normalized across all businesses. This opens up the possibility for a lot of great left hand filtering options. Attribute taxonomies can be presented to the user at this time as well to capture things like hours of operation, methods of payment, and more.

The bottom line is that a local search needs to have a robust vocabulary to support the precision and complete search results that users have been trained to expect from general search engine environments.

WAND's local search taxonomy can augment an existing yellow pages category schema with tens of thousands of additional categories (for query recognition and precision business tagging) and hundreds of thousands of additional keywords (for even more query recognition). WAND has also developed elegant business tagging tools that can easily be integrated into existing business registration or profiling environments.

Please reach out if you'd like to learn more about any of WAND's local search solutions.