During the development of the engine, engineers at New York-based Hakia Inc. had scoured each of those Web pages for what the company founders call "knowledge bits," or pieces of relevant information such as dates, names of people and events to provide more pertinent results than current engines can.
Each page also includes a virtual guide located prominently at the top urging users to ask further questions for more details about the search subject. In this case, the response to the question "What was George Washington's greatest accomplishment?" the guide answers that it was "never losing so great a battle that he couldn't continue to fight." The results also note his accomplishments associated with the Revolutionary War.
The final version of the new search engine is slated to be fully functional by the end of this year. The technology is the brainchild of founder and CEO Riza Berkan, who spent much of his career using artificial intelligence to classify documents related to the country's nuclear weapons programs while working for the U.S. Department of Energy.
Berkan, a nuclear scientist by training, launched Hakia in 2004 with a plan to apply natural language processing techniques in the development of a search engine. Berkan wanted the finished product to interpret the meaning of the information a user is seeking based on the words typed into the search box instead of just indexing pages based on keyword matches, he said
Hakia, Berkan said, is different from today's typical index-based search in two ways.
First, the company's developers have spent the past two years manually building maps of concepts and how they are related so "by looking at the questions [in search queries], we know what concepts are involved." Also, instead of indexing search results based on the words found on a Web page, Hakia uses a technique it calls Qdexing to bring back more relevant search results. Qdexing involves the analysis of individual Web pages for the "knowledge bits," like dates, names, events and other pieces of relevant information to help generate better search results.
"We spend much more time on any given page to understand the concepts involved, the credibility of the page and many things like that," Berkan said. "The number of pages [returned from a query] doesn't really matter much if you think about it. The number of knowledge bits is the important thing."