Hacker News Reader

Comment by _delirium | Hacker News Reader

While I can understand not giving out details, to avoid gaming, this post as a result doesn't have much content! As far as I can tell, this is the entirety of the description of the change:

we developed an algorithmic solution which detects the merchant from the Times article along with hundreds of other merchants that, in our opinion, provide a extremely poor user experience

The explanations of what Google didn't do and why are informative, though.

View on HN

Well, they can't really say so clearly exactly what they did. In an effort to stop people from gaming it, they keep their exact algorithm hidden. When people get around it, they tweak it further, in an endless cycle.

My guess is that their solution isn't actually based on customer's true opinions of the vendor, and thus the solution is cheatable. Hiding the solution is probably their best option at this point.

they keep their exact algorithm hidden

They didn't keep the exact algorithm hidden, they kept the whole algorithm hidden. They didn't even describe it in general.

Maybe the details will appear on Wikileaks?

The people who are trying to game google, know exactly what the solution from google is, or will find out within weeks. The ones left in the dark are the others, that is, the innocent, or the people who are not trying to cheat google, who equally, can not hold google to account.

Thus, their secrecy serves no purpose but to further their own interests.

The "cheaters" will probably have a clue at some point, but unless you are skirting the edge this is unlikely to have much effect on you.

Total guess, but I believe Google licenses Yelp's data. They may be using the Yelp ranking in place of sentiment analysis to reduce the weight of those links.

they just crawl yelp like any other site, yelp happens to use semantic markup (microformats) on its data.

Google does show special search snippets, and _used_ to show yelp review snippets on place pages, thanks to the microformats markup.

I also do not understand the distinction between the unrevealed "algorithmic solution which detects merchants that provide a poor user experience" and the "sentiment analysis" approach that they say they rejected. I don't get how it is possible to algorithmically detect poor user experience without using sentiment analysis! Isn't negative sentiment more or less the same thing as a poor experience?

What I was thinking when I read it, although crawling reviews is another option, both are gameable though. Even a very good algorithm is still going to have the problem of planted reviews, I don't think we are at the stage where an algorithm can decide if a review is genuine.

Maybe there was no change? They just want to discourage people from doing something similar until they come up with a good solution?

Doubtful - "lets release a fake announcement" type ideas I think would get shot down quickly at Google.

Google's "opinion" for better or worse is fact. Let's hope they figured out a way to differentiate between genuine bad reviews and those with underlying motives.

Even more, once you move to anything slightly controversial, an opinion can be genuine, meaning sincere, but still not valid for the person searching ("they sold me the movie I asked for but it had too much violence!").

Yes, just as with voting on HN -- does that downvote mean it's an inane comment, or just that someone disagrees?

Reply on news.ycombinator.com