Smarter isn't better...yet

Admittedly, Powerset’s semantic search boasts some fascinating features. Its representatives claim that it “understands” search terms in natural language. Instead of searching “NHL consecutive games goal streak,” for example, you can ask Powerset “Who holds the NHL record for most consecutive games with a goal?” and get a highly relevant answer.

To show off its effectiveness, Powerset recently demoed its semantic search within Wikipedia’s data set. The results are intriguing; whereas a traditional Wikipedia search pulls up a list of 10 (usually) relevant results to match a keyword search, Powerset, via disambiguation, determines what it is you’re specifically searching for and provides not only the correct entry, but notes and facts on the subject. Impressive stuff, right?

Not according to Wikipedia founder Jimmy Wales. He thinks Powerset's data set is too small to really test the capabilities of semantic search and, as far as he’s concerned, using Powerset on Wikipedia is fixing what ain’t broke. It’s tough to argue with him; Wikipedia searches almost always give users what they want – perhaps even more so than Google or Yahoo searches do. They also hint at being “smarter” than Powerset might have us believe; as users key in their desired search terms, a drop-down bar begins “guessing” at what they’re searching before they hit the “enter” key.

While it’s silly not to consider a search engine that “understands” us an exciting prospect, the effectiveness of existing methods makes me wonder if we “need” semantic search yet. Powerset claims it works best for research, for those not searching for specific items but instead seeking general information on a topic. Well, which types of users are most likely to use search for research and inductive gathering of information? Anyone in the educational field. The last time I checked, they have large internal databases through which they can gather boatloads of literature on their topics of study, be it government documents, journal articles or online writings. In other words, they’re doing just fine. Is there a demand yet for a smarter search?

Also, Powerset says monster search engines like Google get results via “brute force” – saturating results pages with anything remotely relevant – while semantic search retrieves highly targeted information. As Jimmy Wales suggests, however, Powerset is still unproven. What if it can’t answer one of my expertly worded questions? At least with Google results I know I’m not missing out on anything relevant, even if it means wading through some garbage in the process.

For example, I asked Powerset "What is the most poisonous animal on Earth?" and received a detailed report on the Golden Poison Frog (so deadly that other animals have died just from touching a paper towel stepped on by the frog. Who knew?). No other animal was said to be as venomous. When I typed the same question into Google, not only did it also pull up the Golden Poison Frog, it pulled up several other animals who were also thought to be as poisonous. Sure, Powerset's answer might be the correct one, and it may have "understood" the question better than Google, but it's unsettling to see that there are some conflicting opinions on the subject out there and that Powerset missed them.

Powerset also has some kinks to work out with "incorrect queries." Wondering where the Fonz disappeared to, I asked "when did Henry Winkler die?" only to receive a mishmash of results containing Winkler's name loosely associated with the word "die." Typing "Henry Winkler died" into trusty Google, however, pulled up a site that gave me a much more precise answer.

Sure, semantic search may be the way of the future, and it could one day put heat on the likes of Google or at least make them rethink their own search infrastructures. But I don’t think that era is upon us just yet.

I'll stick with the big search engines for now. They may not be “geniuses,” but they’re reliable and smart enough to keep most of us happy.

Posted by Matt Larkin

View Posts by Category

Recent Posts