• http://www.raisemyrank.com/ Bob Gladstein

    Actually, 200 x 50 is a bit more than 1000, but no matter. This reminds me of the days when the home page of each search engine would announce how many billions of URLs they had indexed.

  • http://hernan.amiune.com/ Hernan Amiune

    In fact, in data mining is well known that less dimensions or features lead to a better classification: http://en.wikipedia.org/wiki/Dimension_reduction http://en.wikipedia.org/wiki/Feature_selection

  • awhitfie

    Wow google is Needs to always get a leg up! Your right tho it doesn’t really matter

  • http://www.mattcutts.com/blog/ Matt Cutts

    Hey Danny, this article is actually the opposite of what I was trying to say–I like that Google has stuck with its “200 signals” number, and personally I would prefer not to increase that. I clearly didn’t answer the question yesterday clearly enough, so let me try again here.

    Someone essentially asked “Bing says that they have 1000 signals. Are you going to increase the number of signals that Google says it uses?” And my guess was no.

    The example that I gave was that my team has at least one classifier which in the past has looked at roughly 50 different signals to assess how spammy we think a page might be. We didn’t count those 50 signals as part of the 200 though; we counted the final signal, which was the output of our classifier. Likewise, PageRank is essentially a singular value decomposition over tens of billions with hundreds of billions of links. So we could get into a “signal fight” and claim that Google uses gajillions of signals if we wanted to.

    But we don’t want to. Google’s philosophy has been more about engineers writing understandable and debuggable algorithms based on our experience and ranking intuition. You don’t want thousands and thousands of signals in that sort of scenario. An alternate search engine could use (say) a neural net along with thousands of ranking signals. That might work well 95% of the time, but the other 5% of the time when a query fails and you want to debug it, it’s much harder to debug a neural net than to identify why a more intuitive algorithm didn’t work well.

    So my point was that yes, Google could make claim to using tons of signals in our ranking if we wanted to make that claim, but I didn’t see much value in getting into a “we use more signals than X” debate, and that too many signals can come with negative side effects.

  • http://www.michael-martinez.com/ Michael Martinez

    I use only 1 signal, myself: whether I like what the search engine shows me, whichever search engine that may be.