Directory Search Applications
This class of application includes searching Yellow Pages, online product catalogs, subscription data publications and TV listings.
Search is a critical component within the business model of most directory search applications. Great search directly supports the bottom line in Ecommerce, and creates stickiness where the business model is Web-traffic dependent.
THE DATA LANDSCAPE
A key distinguishing factor for directory search is the nature of the data items to be indexed, which tend to be very short. In a typical enterprise search application, documents will on average be hundreds or thousands of terms in length. Directory data items can be just a few words long.
Directory data also tends to be structured into fields; name, address, phone, email; or color, style, size, price, etc. As with other search systems, most users just want to type a few words into a box, although directory search applications sometimes provide an optional advanced search function enabling the user to be field-specific.
Words from a simple query box need to be matched against multiple fields. However, simply mapping the query to all fields generally creates too much recall and unsatisfactory search results. Directory search users have a preference for precision. A range of strategies can be deployed to improve precision, including automatically "phrasing" search terms.
For example, a user search for red wine would be automatically modified to "red wine".
Where it is impractical to maintain a dictionary of such terms, vocabulary can also be automatically derived from the data set. One approach is to use statistical bi-gram and tri-gram analysis or the data set.
A soft, or "fuzzy" approach to matching multiple search terms to different fields can often pay dividends. Using this approach, the search terms can be made to focus on the most appropriate fields.
For example an eCommerce search for thomas pink, a brand of shirt, could be focused onto the brand/manufacturer field so that shirts by Thomas Pink rank higher than shirts made by Thomas Nash, but which happen to be pink in color. This fuzzy preference for complete idioms is favored by Ecommerce applications in particular. This can be achieved with most search engines by introducing a sophisticated query parser.
Other query related necessities for directory-style search are a good spell-checking / spell-suggestion system, and query auto-completion which not only saves time for the user, but also produces a more detailed query, giving the search engine more to work with.
Many directory search systems also use "Best Bets", or hard-wired results to promote specific items against specific queries or query terms. This can be done to further improve the search experience, or to promote "premium listings", or to publicize items, in context, which provide a high sales margin.
Most systems will also provide results navigation options to help the user quickly refine the results and focus in on wanted items.
Where query volumes are significant, user behaviour analytics can be used to support auto-suggestion of search results. "People who bought this..", etc.
Search engines are also a highly flexible way to proliferate "landing pages" to be indexed by Google, for those wishing to play the SEO long-tail.