Can you give us a quick background of Sphere and how it started?
We originally started with Oddpost, an investment I made with three other founders. We started getting press, particularly from bloggers such as Walt Mossberg, John Battelle, and Dave Winer. Between the three of these guys, and the buzz that they generated, our traffic increased by 23 times. We learned with this that bloggers are VERY powerful, and our question then became how do we get more bloggers to notice us? We started thinking about how blog content is discovered at that point.
Oddpost was soon acquired by Yahoo, and at the same approximate time Google's Gmail was launched. Oddpost was one of the first Web 2.0 companies to be financed and acquired, which was great news. The bad news was that we were just starting to get excited about discovery of blog content, but we weren't able to pursue that with Oddpost.
So, in 2005, Sphere was co-founded with the original founders of OddPost. We acquired venture capital funding, which was half a million initially for our blog search and discovery engine. One of the problems that we were most anxious to solve was the dilemma of blogs on the Web and finding them - it's the proverbial needle in the haystack, which is just exacerbated in the blogosphere since there's just so much content. So we built our algorithm to address that, had a private beta launch, and the results have been very positive.
How does Sphere find and index blogs?
Sphere's algorithm has three basic variables with variables within those variables. We look at link structures - for example, Om Malik is one of the top broadband bloggers. Two hundred other broadband bloggers might link to him, he might link to a couple other people. We can figure out what those other blogs are about, and are they relevant to his topic? Just because he links to you does not mean you are an expert in broadband.
We're also looking at metadata (similar to Technorati), but with more variables, such as depth of the post (it it's only one or two sentences, it might be similar to spammy blogging). This does not imply you need to write a manifesto, but it does potentially flag you. For example, let's say Om's post frequency (first piece of algorithm) was 10 posts on any given day: 7 on broadband, 1 on the Web, 1 on something frivolous. We figure out that he's blogging about broadband (semantically analyzed), we figure out how often he's posting about broadband, which elevates or identifies experts in the category.
The third thing in our algorithm is called "semantic analysic capability" - basically this helps to identify spam (repetitive word patterns), we don't kill spam, but it does get pushed down. This also allows contextual matching, which matches that blog post to images, books, etc.
Go to page two for more of this interview with Sphere co-founder Tony Conrad.
