Google Explores Re-Ranking Search Results Using +1 Button Data

tekgoblin writes "Google plans to use data from its +1 button to re-order search results and keep spammers at bay. While this would bring Google’s search engine into the social networking era, it would also create a new avenue for blackhats to manipulate search results. From the article: '"Google will study the clicks on +1 buttons as a signal that influences the ranking and appearance of websites in search results," a spokesman wrote. "The purpose of any ranking signal is to improve overall search quality. For +1's and other social ranking signals, as with any new ranking signal, we'll be starting carefully and learning how those signals are related to quality."'"

Comments

Well, it's a good idea

[grasshoppa]

However, it's just one more service Google Apps customers ( you know, us paying folks ) can't use.

Only any use if ...

[Stormthirst]

... they also provide a -1 too. I'd like to mod a page down as a method of balancing.

Re:Only any use if ...

[Solandri]

A +/- system suffers the same problem as Slashdot's moderating system. The majority uses the -1 to punish the minority.

Say 4 of 5 people hold a majority view here. Say there are 400 posts representing the majority view, and 100 posts representing the minority view. Say on average there is 1 randomly selected moderator per 10 posters, and the moderators' views have the same distribution. And say that Slashdot only allowed positive mods.

There are 40 mods giving +1 to 400 majority-view posts, for an average of 40/400 = +0.1 per post.
There are 10 mods giving +1 to the 100 minority-view posts, also for an average of +0.1 per post.

Now toss in negative mods. Say one in ten mods gives a -1 to an opposing viewpoint rather than a +1 to their favored viewpoint.

The majority view gets 400 posts, 36 +1 mods, and 1 -1 mod, for an overall average score of 35/400 = +0.0875 per post.
The minority view gets 100 posts, 9 +1 mods, and 4 -1 mods, for an overall average score of 5/100 = +0.05 per post.

The situation gets worse the more people tend to use negative mods. When the ratio of negative to positive mods matches that of the distribution of views (i.e. 1 negative mod for ever 4 positive mods in my example), the negative mods from the majority completely cancel out the positive mods from the minority and the minority view ends up with a 0 ranking average. If the ratio of negative to positive mods is greater than the ratio of minority to majority views, the posts representing the minority view end up with an average negative ranking. Algebraically:

p = % of positive moderations
n = % of negative moderations
A = majority population
B = minority population
Average majority view ranking = Ap - Bn
Average minority view ranking = Bp - An
It's pretty easy to see that if A > B, this skews the majority rankings to be higher than the minority rankings. And if A >> B, B basically has no say in the rankings, and the rankings are almost entirely determined by A's opinions.

So basically negative mods act as a force multiplier, allowing the majority to influence rankings beyond their actual numbers. That is, negative mods tend to produce rankings which reflect the majority view, rather than a utilitarian view. If you want rankings which reflect how useful a site is to the people who want the info on the site you use a +1-only mod system. If you want rankings to reflect the majority's opinion of a site even if it contains nothing they wanted to find, then you use a +/- mod system.

If Google were to allow -1 mods, expect pages for niche topics like Linux to get pounded into the negative, while pages for larger market-share products like Windows are barely affected. Basically, hundreds of millions of people would do searches on Windows topics, and -1 the occasional Linux page which got into their search results. But those "occasional" -1s against Linux pages would likely far outnumber the +1s given to Linux pages by the few million people doing searches for Linux.

Google tricks

[Clsid]

I'm truly impressed at Google's ability to makes us work for them without most people even noticing it. This is yet another example of that. Then I hear non-tech people in awe of how smart Google is, and then say to myself, that is one hell of a business model.

Re:Google tricks

[ajyasgar]

It really is one hell of a business model. They use us as essentially unpaid employees, and we continue to use their services every day. Our use of their services, our very existence, provides them with valuable and necessary data to continue with their business plan/model. This is effectively a mutualist symbiotic relationship - both parties are benefiting greatly from this, the only difference between this one and the ones we commonly see in the animal kingdom is that the anemone us clown fish live in protects almost the entire population of clown fish simultaneously, while allowing all individual clown fish access to any information generated by groups of the others, in addition to any populations living outside the anemone. I don't see a problem here. The uproar over privacy starts anytime Google is mentioned but really, people are more than willing to provide the Internet - anyone who wants it - with their close personal information. For now, Google is still doing us a favor.

Re:Google tricks

[Raenex]

Yeah, they still pump your gas for you in New Jersey.

It's actually required by state law. Personally, I prefer self-service.

"Proposed", not "plans to"

[Sockatume]

It's a research project into how useful this metric would be. The article makes it clear that they haven't made any plans to actually implement it until they know how well it works. At this stage, blackhat activity would lead that project to a negative conclusion, and the feature wouldn't be implemented, and it couldn't be manipulated.

Re:"Proposed", not "plans to"

[mfh]

At this stage, blackhat activity would lead that project to a negative conclusion, and the feature wouldn't be implemented, and it couldn't be manipulated.

Unless the blackhats were smart, which most are. In which case they will use this testing stage as a way to elicit false confidence in this new avenue while gathering information they will need to make the new changes work in their favour. Black hats almost never do what you'd expect them to do. They often do the exact opposite until just the very proper moment.

Re:"Proposed", not "plans to"

[ginbot462]

My exact sentiment. You would either "play nice" or not play at all until it goes live. Even then ... you might wait for the PUBLIC to gain confidence, then launch what ever human/computer methodology to pump up your desired links.

Re:This is a really bad idea

[JanneM]

This is about the +1 button on search results, not G+. G+ has nothing to do with this.

There is a risk of spamming, but it depends also on how they end up using it. If, say, it ends up promoting things similar to those I've +1'd, but only for my own searches, then the risk of spamming is quite low. If similarity is determined by the +1 from people that tend to +1 similar sites to myself then you can't really spam it at large scale.

A user that +1s every site in sight will not have a similar profile to anybody else but other spammers. If they +1 their own sites and a coherent set of other sites then they'll only target the specific people interested in those kind of sites; too much work for too little effect for it to be an effective manipulation method.

Leveraging search to get identity entries

[David Gerard]

Better than that - they're actively leveraging the search engine to collect identity profiles. Join G+ or your page falls down the search.

They're actually willing to compromise the search to collect profiles.

Re:This is a really bad idea

[robthebloke]

Once 4chan gets wind, it's safe to assume all google searches will return rick astley....

Re:This is a really bad idea

[m.ducharme]

Which may be why Google is pushing so strongly (rightly or wrongly) for verified identities on G+.

Re:This is a really bad idea

[Abstrackt]

A user that +1s every site in sight will not have a similar profile to anybody else but other spammers. If they +1 their own sites and a coherent set of other sites then they'll only target the specific people interested in those kind of sites; too much work for too little effect for it to be an effective manipulation method.

Just out of curiosity, what prevents spammers from +1ing a coherent set of sites plus the one(s) they're trying to push? For example, what if you have a spambot advertising dating sites that also +1s a bunch of tech sites (i.e. Slashdot)? Wouldn't that effectively create targeted advertising?

Re:This is a really bad idea

[m.ducharme]

...if we're lucky.

Re:This is a really bad idea

[aeortiz]

I think this is why they're insisting on real names for Google+.

When every user is a person, it is more difficult for spam CEO hackers to skew Google results. I suppose they could still try to harvest millions of Google accounts to use as +1 slaves but that's a lot harder than setting up a content farm. Spammers will have to create fake personas or steal real ones in the millions to be able to cheat now.