I would like to have a search engine that has the ability to re-tune itself based on my behavior. It would build a history of the pages I select and note patterns of information within sets of related pages. It would use that knowledge to construct more refined queries.
I propose a search engine that is, for lack of a better word, teachable.
People have been thinking very hard about customization and relevance in search results for at least a decade. And some of this sort of technology already exists, hidden deep within large search systems.
For a look at a closed-knowledge universe–which can also be considered a database of library holdings–that has some characteristics of teachability, consider Kanisa’s ServiceWeb system (www.kanisa.com) for online customer-service information.
The system, used by large corporations including Microsoft, Adobe and Apple, takes a vast database of customer-service answers and arranges it into a hierarchical map.
When a customer submits a query, Kanisa’s system dissects the question and puts it in context on the map. Then, the front-end engages the user in a dialogue to refine the query, drawing finer and finer distinctions until it arrives at the desired piece of knowledge.
This is close to what I want–but no cigar. Kanisa’s system demands that the user actively say what he or she wants, whereas I envision a system that can learn by simply observing user behavior. Also, Kanisa’s system is not cumulative, in that query precision does not improve with more queries. In that sense, it is guidable, but not teachable.
Another Web search technology that was both cumulative and built on passive observation was Direct Hit. That system, which was used in search products by everyone from MSN to Lycos at one time or another, achieved greater relevance with a “popularity engine” and a database of results from millions of searches.
Direct Hit looked at what people actually clicked on and how much time they spent on those sites. But the company was never able to get portal companies interested in developing personalized search technologies.
Direct Hit was swallowed by Ask Jeeves in January 2000. Today, its relevance-ranking technology is used by Ask Jeeves in search services for third-party Web sites. The Jeeves folks have a separate Web search product geared directly toward consumers, Teoma.
The 2-month-old search engine (www.teoma.com) features a dynamic technique for improving relevance that’s somewhat similar to what Kanisa is doing. Teoma looks at clusters of similar sites on the Web and allows users to make fine distinctions about sites within those clusters. Start, hypothetically, with a search for “wine,” then “wine producers,” then “Napa Valley wine producers.”
There’s also Outride, the Redwood City start-up that was acquired by Google last September. Outride, a spinoff from the fabled Xerox Palo Alto Research Center, had a “relevancy engine” that looked at a lot of user behavior, including navigation, search history and time spent on a given site.
More Top Picks Best Rtx 4070 Graphics Cards
According to former Outride CEO E. Casey Roche, a “100 percent increase in user productivity in searching was achieved very early on. For example, a car enthusiast searching for `Escort’ got the Ford Escort and avoided pages of ads for escort services.”
Google has yet to deploy the Outride technology.
For many companies that may be toying with taking these technologies to the next level, I believe the bottom line is the trade-off between privacy and relevance. How much about our tastes and ourselves are we willing to furnish to search companies in the name of enhanced performance?
Ranjit Padmanabhan, a reader, suggests we’re close to the day when search engines will offer enhanced service on an informed consent model similar to what Amazon uses for recommendations.
“I predict that by Christmas a company that commands a large audience will offer personalized search as an option for users in exchange for registration and the willingness to have their search activities monitored and measured,” he said.