When you go to search the web, is it because you want to read through a lengthy article related to a subject, or do you just want the facts that answer your question? Factery Labs, a new service that’s launching this morning, is hoping it’s the latter. Factery is a new search engine/API that uses advanced language processing to sift through content on the web to identify the most factual statements — in other words, it takes news articles and webpages and breaks them down into a handful of bulletpoints on the fly.
To give an idea of how Factery works, the company put together a simple search engine with a two-column view: one does a search for your query on Twitter and parses facts from any articles linked from those matches; the other column uses Yahoo’s BOSS engine to look at articles that are less time sensitive. In this context, Factery has its hits and misses. The site stumbled on “Danville, CA” (a city in the East Bay), yielding very few results for both Twitter and BOSS. Other times it fared very well: for “Arc de Triomphe” it generated quite a few interesting facts (you can see a few in the screenshot) though it sometime seemed to grab all of its Facts from either Wikipedia or Answers.com — it seems like it would have just been easier to read the Wikipedia article itself.
As for the Twitter integration, I found the results to be pretty poor for queries that weren’t about breaking news. The Arc de Triomphe example worked well for the Yahoo results, but the ‘facts’ from Twitter were useless, with non-sensical results like “The image is about 8×10”. But for queries related to breaking news it worked well. A search for “Leonid” (as in the meteor shower taking place this morning) showed very good results for both Twitter and Yahoo.
But the performance of this search engine isn’t especially important, because Factery isn’t setting out to be a search engine destination — at least, not yet. At this point the service is looking to offer its API to developers, which is where Factery’s real potential lies. One obvious use case is in Twitter clients. Any time someone shares a link on Twitter, there’s typically very little room for them to give it any context — oftentimes you’re left with a recommendation that you should check something out, with no idea if you’re really interested in the linked article. But if your Twitter client has Factery integrated, it can present a few bullets summarizing the article before you click it. One web client called Sobees already has it implemented, and the company says it is in talks with others. There are numerous other obvious uses for this on the web. For example, sites like Digg or Topsy could use this to do a better job describing linked articles to users.
Factery is making many of the same promises made by Powerset, the semantic search engine that was acquired by Microsoft in summer 2008. This is a tough problem, and at this point I’m not convinced Factery would garner much popularity as a standalone search engine for two reasons: for one, it isn’t consistent enough that I’d choose it over Google when I was in a bind. And two, the ‘fact engine’ idea is a big departure from the ‘link engines’ we’ve spent a decade using religiously, so it will take a while to get used to. That said, its future as an API looks bright, especially if it can land some of the bigger Twitter clients.