As of January, 2012, this site is no longer being updated, due to work and health issues
Search engines use statistical and lexical analysis to match query terms to indexed text, but often, human judgment is more effective. Many search queries are for names, items or even ID numbers. For the most frequent queries, manually identifying the best page or pages, and presenting these the search results as suggestions avoids any need to tweak the algorithm, while still generating excellent results.
Adding human understanding to makes the results more useful for the common cases, while the search engine's dynamic retrieval and relevance functions handle the unique and unusual search terms.
Eight Principles for Good Search Suggestions, analysis by Avi Rappoport, SearchTools.com
Dell does a pretty good job of this: they make sure that a bunch of appropriate introductory pages show up when users type the query "linux" on their site. As shown the example below, if they had not done that, the search engine would have shown them only the Large Business Linux pages first, and people in other situations would have thought that there was no Linux solution for them.
Another way to approach this is to consider the most common searches on a site as candidates for a Knowledge Base, and the manual links in the search results as pointers into that database or FAQ.
The 2001 Forrester Search Report quotes a telecom company site administrator as saying "About half of our visitors come to the site looking for a specific product." When a site visitor searches by a code, such as an error message or product ID, the search engine should recognize the pattern and make sure that the appropriate types of pages, such as product specs, troubleshooting or FAQs, come first in the search results.
For example, at the Sharp USA site, searching for a laser printer using Atomz search and promote tools brings up a special offer:
Articles on This Topic
- Eight Principals for Good Search Suggestions, SearchTools.com, January 4 2007, by Avi Rappoport
A practical approach to making best use of resources when creating and maintaining search suggestions.
- How new site search led Maps.com to 20% hike in sales Internet Retailer, October 28 2004
Implementing Atomz Search allowed Maps.com to recommend the most successful world maps (those with the lowest return rates) in search results.This lead to significant improvements in sales in July over June 2004.
- Taking E-Commerce to the Next Level. Business Week, August 31, 2004, by Amey Stone
Describes retailers using search suggestions to highlight special promotions.
- Intranet Usability: The Trillion-Dollar Question Alertbox at Useit.com, November 11 2002 by Jakob Nielsen
Results of an international usability study by the NNGroup on intranets finds that many are wasting employee time by failing to provide usable intranets. They found that "search usability accounted for an estimated 43% of the difference in employee productivity between intranets with high and low usability." They recommend that intranets make sure that the main search engine indexes all pages, shows results in relevance order with search suggestions at the top, encourage useful page titles and descriptions. The NNGroup finds that tasks requiring 27 hours annually on a usable intranet could take as long as 196 hours on a less-usable one.
- Beyond the spider : the accidental thesaurus. Searcher Magazine, October, 2002, by Rich Wiggins
An excellent article providing examples of search suggestions at the AT&T web site and AOL's Keywords. A case study of creating the Keywords feature at Michigan State University describes the process of designing and maintaining a search suggestions database. Also analyzes the distribution of queries in the search logs, which conform to the Laws of Pareto, Bradford and Zipf, showing that concentrating on a very few common cases will provide the best return for effort. A similar function, implemented at Bristol-Meyers Squibb, has had great success.
- Why search is not a technology problem: Case Study, BBCi Search (follow link to 3.4 MB PowerPoint file) ASIST IA Summit February 2002 by Matt Jones
Describes how the British Broadcasting Corporation added a search engine to their web site, offering access to both all the information in the BBC site and web-wide information. The group created some "use models" to show examples of information needs and find ways to add context to search results. They chose to create a taxonomy based on the most frequent searches, and test the process over the course of several months, starting with paper prototypes. They found significant user sensitivity to wording and layout, and to functions that break with their expectations. Showing search suggestions was successful once they integrated with the other results, and people loved the search zones tabs once they found them (after several visits). Recommends early testing and use models to learn how people really use search engines.
- New HP Wetware Product leads to Smarter Search Louis Rosenfeld's Blog, September 13, 2001
Describes the suggestions links at the Hewlett Packard site for such common topics as handhelds, and suggests that the hybrid of search "best bets" for popular queries with standard search results is a hopeful trend.
Examples of Sites with Suggestions
- BBCi (search for castles, look for "BBC Best Link" and to the right for "BBC Recommended" sources)
- University of California, Berkeley (search for human resources)
- Microsoft (search for Bill Gates or search engine, top results are also linked on the right)
- Michigan State University (search for zoology or library)
Comment on this article