Hacker News Reader

The thing that bothers me about all this drama is that the actual offense Google wants everyone to be so worked up about is that Bing doesn't filter Google from its clickstream data.

Bing wrote code that works across the whole web. The whole web includes Google. As a result, Bing gets some info from Google. But they didn't get that info because they copied Google, they got it because they didn't filter Google out - or, said another way, because they ignored Google as someone they needed to special case for clickstream analysis.

I don't work in search, but the idea that you're supposed to special-case your competitors when writing general-purpose tools sounds an awful lot like a unilaterally recognized gentleman's agreement. If it's not illegal, and it doesn't hurt end users, why shouldn't it be considered fair game?

I also think it's odd that throughout this whole thing, nobody has really noticed that the only possible way Google could have spotted this issue is if they're keeping very close tabs on Bing's search results. It's another arbitrary line that Google seems to have unilaterally drawn: it's clearly fine to monitor your competitors results closely, which presumably is going to have an effect on your own results; it's only out of bounds when that effect is directly measurable.

I thought part of the point is that whatever Bing is doing doesn't work across the whole web. They need to associate the URL with a query, and most websites don't have queries. It's not just that Bing has recorded a click on Miley Cyrus's webpage; it's that they've done that and associated it with the query [kecgxjpgqoe].

People have made that point, but I don't understand it. 'kecgx...' shows up as a parameter in the url of the Google query. Tons of sites include relevant information in parameterized urls; why is it unexpected that Bing would use that information across the whole web? Other people have said that implies that Bing has to have special Google url-parsing code, but that's not true at all - query parameters in urls are standardized. You would have to have special code to understand the specific semantics of Google's query urls, but there's no reason to think Bing needs or wants parameter semantics, they could easily just be interested in making probabilistic associations.

"You would have to have special code to understand the specific semantics of Google's query urls"

That's why people say that it's special-cased. There's no web standard that says the 'q' parameter means that the page is a search engine and the parameter is the query. That was something AltaVista did a long time back (possibly for byte-saving reasons, or possibly because they were lazy) and Google et al copied. Many other search engines use a different system, eg. DDG puts it in the request path, InfoSeek used qt=, Excite used search=.

Why do you think Bing cares or has to know that "q=" means a Google query term? My point is that they don't have to have any semantic information about parameter keys to be able to derive probabilistic associations between parameter values and clicks.

If you consistently see pages with 'foo' as a parameter value to any parameter key, and clicks on those pages consistently go to site bar, it's completely reasonable to start associating foo with bar, regardless of what the parameter keys are.

> Why do you think Bing cares or has to know that "q=" means a Google query term?

If that's true, then they should also be associating the sites linked with all the other weird parameter values in a search query, which would spam them to heck. Here are all the params from a search I just did on google:

q=test&ie=utf-8&oe=utf-8&aq=t&rls={moz:distributionID}:{moz:locale}:{moz:official}&client=firefox

They'd start making a lot of strange associations between random sites and "utf-8" if that were true, because that parameter shows up in just about every Google search done in English.

It's also a perfectly normal thing for programmers to search for, so they'd clutter up their index with millions of sites that had nothing whatsoever to do with utf-8.

So to make any real use out of that, they had to understand what the parameters in there actually mean, rather than associating ALL of them with whatever site was next in the clickstream.

Though I grant you, that does not disprove the alternate hypothesis that they were dumb and polluted their index with loads of irrelevant crap.

And I admit that I found weak support for that hypothesis by trying to see if utf-8 was linked to rim.com (one of the tests sites, if memory serves):

http://www.bing.com/search?q=utf-8+rim&go=&form=QBRE

Those results appear to be crap, though I'm not sure that any sensible results exist that they could return.

If that's true, then they should also be associating the sites linked with all the other weird parameter values in a search query

If it's a parameter value people will ever actually use Bing to search for ('utf-8'), there's probably plenty of other signal to help them figure out which results to return. If it's not ('kecgxjpgqoe'), we already know they sometimes return crap, thanks to Google's little experiment.

They'd start making a lot of strange associations between random sites and "utf-8" if that were true, because that parameter shows up in just about every Google search done in English.

If it shows up as a parameter for every search, why do you think Bing's algorithm would decide it was a good indicator for a particular url? P(foo.com|utf-8) wouldn't be any different from P(bar.com|utf-8), making 'utf-8' basically worthless as a discriminator. I'm pretty sure the folks at Bing understand the concept of conditional probability.

It's also a perfectly normal thing for programmers to search for, so they'd clutter up their index with millions of sites that had nothing whatsoever to do with utf-8

I see no justification for the idea that url parameters are somehow a more difficult challenge in this regard than the mass of crap that is content on the web, which we know they crawl and index at a massive scale.

It would only show up as a parameter "relevant" to whatever sites were next in the clickstream. Remember, not every page transition is Google -> other site.

They'd also be gobbling up tons of random things from forums and whatnot (which should appear on long tail searches, if we knew where to look), most of which spam the heck out of you with random parameters, forum names, and whatnot.

> I see no justification for the idea that url parameters are somehow a more difficult challenge in this regard than the mass of crap that is content on the web, which we know they crawl and index at a massive scale.

Which is why I see no reason to assume that they don't understand or attempt to understand the actual meaning of parameters passed to one of the biggest sites on the internet.

You said 'utf-8' shows up as a parameter in almost every English Google search, and suggested that this would cause weird associations between 'utf-8' and random pages.

I pointed out that there's no reason to expect this to be true, because Bing engineers are likely smart enough to realize that if the probability of clicking on to foo.com is not significantly different than the probability of clicking on to bar.com given the presence of the 'utf-8' parameter, then 'utf-8' is a pretty poor discriminator between foo.com and bar.com, and probably shouldn't be used to determine search results.

It doesn't matter that not every page transition is Google -> another site. You still wouldn't need to special-case Google to determine an association with a parameter is useful or useless - the same code could build that model for any site with params.

They'd also be gobbling up tons of random things from forums and whatnot (which should appear on long tail searches, if we knew where to look), most of which spam the heck out of you with random parameters, forum names, and whatnot.

How do you know that they don't? Google pointed out some longtail results that look bad, and you yourself pointed some out in a previous comment.

Which is why I see no reason to assume that they don't understand or attempt to understand the actual meaning of parameters passed to one of the biggest sites on the internet.

You're misunderstanding what I'm saying. I don't assume that they don't; I'm just saying there's no evidence that they do, and that assertions of wrongdoing based on the belief that they do are just irresponsible speculation. I have no knowledge of what Bing actually does, but neither do the vast majority of the people on the internet who are talking about this, many of whom assume the worst based on a mistaken notion of what's technically necessary to see the results that Google demonstrated.

> the presence of the 'utf-8' parameter, then 'utf-8' is a pretty poor discriminator between foo.com and bar.com, and probably shouldn't be used to determine search results.

And yet, it will link random text to websites even if they appear only in Google's URLs. I realize you're talking about discrimination (as in, "what's the better result for utf-8?"), but if the code is generic, it ought to be generic in this respect as well. After all, it linked up random nonsense to random sites given nothing more than Google's say-so, even though there's plenty of information out there about, say, rim.com that would tend to indicate that nobody except Google thinks that random text is relevant to an otherwise well-known site.

> How do you know that they don't? Google pointed out some longtail results that look bad, and you yourself pointed some out in a previous comment.

Indeed, I do not know. I know that it would be dumb to link those things to random sites, but you are correct that I do not know if they're doing things that dumb.

> I don't assume that they don't; I'm just saying there's no evidence that they do, and that assertions of wrongdoing based on the belief that they do are just irresponsible speculation.

Well, for one, I'm not really asserting "wrongdoing" here. That is, I don't particularly think that it's wrong of them to do things this way. My interest is mainly technical, so I'm more interested in figuring out exactly what they're doing rather than blaming them for it. As such, I'm going for the most likely explanations I can find, rather than worrying about whether it's been proven to such an extent that they can be blamed for it (as I'm not really going to blame them anyhow).

You may have seen where I pointed out that I don't think it will "ruin search" in the end because they should expect a crapflood from spammers now that it's clear that they use clickstream data to rank sites. After all, there's a large spam attack right now on a tiny wiki for a game I play. I have to think Bing is more of a target than that. I can't prove that, true, I'm just playing the odds here.

By that standard, Google should also quit trying to integrate invite-your-fb-friends feature, right? Cuz it's essentially facebook who has figured out how to get massive user signups with their real data and it's google trying to leech.

Instead of not trying to integrate fb, Google is accusing fb of not opening up the data. So when it's convenient, you want data opened up. When it's not convenient, you scream "copycat!".

Google was freely sharing contact data with Facebook, so there's no hypocrisy here. Contact data is not the same as a search engine.

Freely sharing the data in the sphere where you're trying to catch up, while zealously guarding the data in the sphere where you've got a massive lead, is the alleged hypocrisy.

Hard-won bits is hard-won bits, at a suitably abstract level of analysis.

Contact data is not the same as a search engine.

Elaborate? Bing didn't steal Google's source code. They intercepted some linkstream data with the user's permission.

A search engine performs the task of narrowing 1 trillion URLs into the 10 that are most likely to be what you are looking for. Associating a query with those 10 results takes teams of software engineers and datacenters full of computers. A user could not reasonably do this on their own.

Contact data is a list of the people you know. I had that list in my brain before I ever visited Facebook. If I am on different social networking sites, that list is mostly the same, because it is information about me that I already knew. I didn't come to Facebook saying "hey Facebook, find for me the 10 people in the world who would be the best possible friends for me." Instead, I come to Facebook and start telling them the names of my friends.

But forget all that for a second. Imagine that contact data and search engines really were the same thing. Google does not access Facebook's social graph, because Facebook has not allowed them to. For Google's behavior to be like Microsoft's Google would have to work around that by using the "clickstream" of Chrome users who are on Facebook. Who knows, maybe Microsoft is doing this already, since they seem to think that "clickstream" is a valid way to mine the internal databases of other companies.

A search engine performs the task of narrowing 1 trillion URLs into the 10 that are most likely to be what you are looking for. Associating a query with those 10 results takes teams of software engineers and datacenters full of computers. A user could not reasonably do this on their own.

Quite irrelevant to the discussion here. A user could suggest specific pages be indexed by Bing, no? If so, why can't a user let Bing automatically listen in on the pages they are visiting and index them if Bing wishes to? That's what Bing did.

Contact data is a list of the people you know. I had that list in my brain before I ever visited Facebook. If I am on different social networking sites, that list is mostly the same, because it is information about me that I already knew.

Again, pretty irrelevant. If we are talking about data ownership, both your contacts data and your clickstream data belong to you. You can do what you want with it. Some users choose to share it with Bing. That's their prerogative. I don't think even Google disagrees with that.

They just don't see the hypocrisy between this and their stance on Facebook.

Google does not access Facebook's social graph, because Facebook has not allowed them to.

Correct, and yet, Google's stance in the Facebook episode has been that Facebook should open up that data because it does not belong them, it belongs to the users. To be consistent with their stance on Bing, Google should never have created a big fuss about Facebook not opening up the data.

Similarly, clickstream data does not belong to Google. It belongs to the users.

Even if absolutely everything you say is true, even if "clickstream data belongs to the users" and the users can "share" it with Microsoft, it's still the case that Microsoft is riding on the backs of Google's engineers.

And as I said in my original message, maybe that's even a smart business move. But let's call a spade a spade. Microsoft is copying Google search results.

Is that really what you want to be sticking up for?

Microsoft is copying Google search results.

Clickstream data != copying search results.

This is the lie that you keep telling yourselves. Rationalize it however you want, but show a normal person the screenshots on this page and ask them if there is copying going on: http://www.mattcutts.com/blog/google-bing/

Maybe you call it "clickstream," maybe you think this has something to do with the user (even though the users had no idea they were aiding in this), this web of rationalizations cannot get around this fact:

Bing puts things in its search index that came directly from Google's algorithm.

...that came directly from Google's algorithm

That is a lie.

1. User searches for specific string 2. Google pulls up the results 3. User clicks on the result

Bing makes a corelation between the user's search string and the URL they end up on and takes a note of it. There is very little algorithm involved here because there is only one listing for that string. The algorithm is primarily used to rank pages. In order for Google to convince me that Bing is copying them, they would need to show a consistent before and after of Bing copying the search results and its order. Emphasis on resultS--in plural.

Neither are present in the Google honeypot. There is only one search result returned; and there is no question of the order of search results cuz of that. Two items central to the algorithm are entirely missing.

Going back to what Microsoft does copy:

1. it copies the user's input--imo this belongs to the user

2. it does not copy the order of search results nor scrape the returned search results returned by google

3. it copies the url the user ends up on, which is effectively a user's browsing history that the user has opted in to share with Bing.

I think you're trying too hard to rationalize Google's claims.

With regard to the 1-in-1000 signals, clearly they have that many signals because not all 1000 signals are strong for every query. In Google's case, they "Googlebombed" Bing, but instead of using the "anchor" signal (which most people use to alter Google's results), they chose the "click stream" signal. Google specifically chose scenarios where the other 999 signals weren't being used.

If I had a toolbar installed on millions of machines where people opted-in to send me their click activity, I could start a search engine that used only click stream data to rank results. You'd be seeing the ranking effected by all sorts of sites including Facebook, Google, Amazon, Wikipedia, etc... As it should, the ranking would be effected by the click activity that users perform on the most popular websites. Effectively I'd have crowd sourced my ranking. If Google is returning a result for a particular person's name, but Facebook's click stream activity is returning a fan page for that person's name which is clicked on more frequently than the Google result, then the Facebook result would show up in my search engine for that person's name.

All Google did here was choose a scenario where every other signal wasn't being used by Bing, effectively turning Bing into the click-stream search engine described above. Furthermore, Google chose a scenario where even within this single signal, they were the only input to it. All Google did was googlebomb Bing. Any well trafficked website could do the same.

It's ridiculous the lengths people will go to while trying to discredit Microsoft. Bing has built a generalized system that partially learns rankings through behaviors observed on high-reputation websites. That's it.

And no they shouldn't remove Google from the signal because that'd imply they've done something wrong.

>As it should, the ranking would be effected by the click activity that users perform on the most popular websites. Effectively I'd have crowd sourced my ranking.

Not exactly. Those sites work hard to present relevant links to the crowd. If the sites have competent engineers, your record of clicks cannot add much value -- if just because you only have a subset of all the clicks on that site. So you see google says they're seeing increasing overlap with bing over time.

Also see http://news.ycombinator.com/item?id=2184025

"The bottom line is that Bing's index contains associations that it could never have figured out if Google hadn't figured them out first."

Actually, it is the users who are figuring out what is relevant not Google since a) Bing are collecting data on the links they click regardless of where Google ranks it (could be on the 20th page of Google search results for all Bing care) and b) Bing are collecting this data from all sites, there has been no evidence that Google search results are treated any differently to any other site.

> Actually, it is the users who are figuring out what is relevant not Google

When you do a Google search, Google chooses ten URLs from over 1 trillion that are in its index. Creating this list takes teams of full-time engineers and data-centers full of servers that crawl the internet constantly.

The user spends, on average, less than ten seconds choosing 1 out of those 10 URLs. In a small minority of cases, they'll click to further pages, in which case maybe they were choosing 1 out of 100.

So Google narrowed the search down from 1 trillion URLs to 10, and the user narrowed it down from 10 URLs to 1. And yet you think it's the user who did the hard work?

Individually, it negligible. Cumulatively, the time spent by the many thousands (millions?) of users who are clicking links they deem relevant for queries all over the web (Bing are collecting data from all sites) is huge...

My point is that the assertion that Google are doing all the work in creating the data used by Bing is clearly false.

> My point is that the assertion that Google are doing all the work in creating the data used by Bing is clearly false.

Clearly it's not 100%, but what percentage is it?

The user ends up visiting 1 out 1T possible URLs.

Google narrowed it down from 1T to 10 URLs (11 orders of magnitude).

The user narrowed it down from 10 to 1 URLs (1 order of magnitude).

Even if you measure this logarithmically, Google did 91% of the work, and the user did 9%. And the Microsoft people are arguing that it's totally justified because of that 9%.

Or you could measure it linearly, and argue that Google did 99.999999999% of the work because that's the percentage of the results they eliminated.

Either way, this is an extremely weak justification for taking a result of which 91% came from your direct competitor (when measured in the most charitable way).

those millions of users are all searching on platforms where 99%+ of the work was already done by the site's own engineers. Cmon, this isn't rocket science.

I can not agree more with your second paragraph.

Yes, Bing toolbar just rely on people clicking links. But without Google doing all the hardwork here, users will never know about such link, that's why they went to Google. And there will not be any of such link to feed to Bing's system.

You could also argue that the user is doing the hardwork by going to Google (or any other site, since this is not specific to Google) and deciding which links are relevant by clicking on them. The user has opted share this work they have done with Microsoft (by installing the toolbar and agreeing to the terms) to help them get better results on Bing.