SAN FRANCISCO (05/08/2000) - If only your search engine had a brain.
DolphinSearch may be the closest thing. The new search engine claims to return more relevant results because it can recognize the actual content of documents.
Its technology incorporates a model of pattern recognition used in dolphins' brains to make sense of "echo-location" data.
DolphinSearch relies on neural networks, which are a type of artificial intelligence that mimics the electrical patterns in brains, according to the search engine's developer, also called DolphinSearch. This technology helps the search engine "read" documents to create a bias or viewpoint in a given topic, and focus its searches of other documents.
For example, the company fed the search engine hundreds of articles from golf magazines to create the golf-specific demo on its Web site.
So, when you enter the query "putter," the search engine responds with documents about putting techniques and putters that pros use in golf tournaments. It produces fewer of the advertisements and irrelevant hits of generic search engines, says Andy Kraftsow, DolphinSearch's chief executive.
"DolphinSearch only knows about putting from the point of view of the golfing magazines that it trained from," Kraftsow says.
The first product will be KnowledgeBox, a networked device that companies can use on their own intranets, as well as on the Internet as a whole. Once KnowledgeBox has read documents on the corporate network, it knows what types of documents would be most relevant to return in an Internet search.
"It's almost identical to that smart person who has read all those things and knows what's in them," Kraftsow says.
KnowledgeBox will ship by midsummer and cost around $10,000.
DolphinSearch probably won't compete directly with AltaVista, Ask Jeeves, or Yahoo in the search engine sweepstakes. Kraftsow says the types of broad, ambiguous searches popularized by those engines aren't the best use of DolphinSearch's technology.
Put to the Test
Does DolphinSearch work as advertised? Rigorous, independent test results aren't yet available. But Henry Lieberman, a research scientist at The Media Lab at the Massachusetts Institute of Technology, says DolphinSearch "seemed to do better, on average," when he tried the golf demo.
"My technical opinion is it's sound, reasonable stuff. I wouldn't say it's earth-shattering. I think the theory of it is, in general, good," Lieberman says.
DolphinSearch hired Lieberman for technical advice, but the researcher is not endorsing the product.
Most other search engines recognize the problem of narrowing down results, Lieberman says. They do statistical analyses of popularity, count the frequency of words, or rely on analysis by human experts for relevance.
DolphinSearch may face direct competition from a technology called latent semantic indexing, Lieberman adds. Autonomy, a British company, categorizes material using neural networks, pattern matching, and a technique called Bayesian analysis.
Another challenge comes from the tiny minority of search-engine users who can confidently wield logical terms like AND, OR, and NOT when entering queries.
"If you know how to use Boolean logic really well, you can get similar results to what we get," Kraftsow acknowledges."