Google's Panda Update: An Interview with Matt Cutts & Amit Singhal | Hacker News Reader

Google's Panda Update: An Interview with Matt Cutts & Amit Singhal

wired.com

28 pointsilldave8 comments

"We’re lucky to have the criticism, because that means people care enough to tell us what they want."

That's a winning attitude in a competitive business. Because the Google team mentioned search results for medical terms in the interview, I tried out a few searches on medical terms that a few months put appallingly bad sites at or near the top of the results page. Now Google is serving up much better sites on medical topics, so I'm glad that algorithm change was made.

There are lots of very interesting things in this post - one of the most interesting long term changes I see is that Google are now admitting that you can lobby for algorithm changes.

I can imagine in another 12 months there will be vocal lobby groups to get Google to change parts of their algorithm for certain types of sites. This feels like the start of Google coming out of their shell and admitting that they listen to communities like HN and will respond with large scale algorithm updates.

I'd also dearly dearly love to know what this breakthrough was:

"Amit Singhal: Well, we named it internally after an engineer, and his name is Panda. So internally we called a big Panda. He was one of the key guys. He basically came up with the breakthrough a few months back that made it possible."

Indeed. My mind races:

something like assigning super-distinctive words prime numbers and multiplying them in a paragraph, and doing a lookup on it to see if that combination of super-distinctive words appears elsewhere.

Would be fun to find out how very wrong I am...

That algorithm would find duplicate content, which Google has done a decent job detecting for a while.

The new update was about low-quality content. Thinking of an algorithmic way to detect that isn't easy (obviously!). I guess I'd be looking at adapting things like http://en.wikipedia.org/wiki/Flesch%E2%80%93Kincaid_readabil... somehow, so it wouldn't penalize either easy-to-read or difficult-to-read but would look at the density of harder-to-read content vs easier-to-read fluff. Or something!

I'm not sure it would be that hard for Google to identify the crap sites - when you go to one you either click your back button or out through the adsense that usually dominates/perfectly blends with the content anyway, Google is catching that bounce rate damn near 100% of the time.

The article is actually quite forthright. Singhal implies that the core of this uses some form of machine learning. Looks like a binary good-bad classifier that can give confidence or a probability in its predictions. Negative ratings have a higher threshold. I'm thinking what is more important than classifier is what features are chosen. I bet the breakthrough was a clever way to choose or identify these features.

It's also interesting how they admit that the results can be gamed if you know what the changes they've made are. While that's most likely always been the case with Google's algorithm (any search algorithm, really), it's interesting that they're publicly saying it.

How long do you think it'll take for people to figure out how to game this change? If content farms are dead, what's next? Social Spam?

Reply on news.ycombinator.com