Hacker News new | past | comments | ask | show | jobs | submit login
Google Correlate by Drawing (google.com)
475 points by franze on Sept 2, 2011 | hide | past | web | favorite | 77 comments

Interesting to see what is dying: English Chinese Dictionaries, radio stations, Altavista, lyrics, text messaging, Flash.


And look what's getting more popular: Porn video sites, video sites, Facebook, Mac computers and how to work them -- particuarly how to take screenshots: this should be a tip-off to those of you working at Apple doing UX. ;)


Some of the stuff you would guess was becoming less or more popular but there is other stuff which is less obvious.

edit: This is super interesting: http://www.google.com/trends/correlate/search?e=id:SnghYCuBW...

Looks like Hacker News is gaining popularity exponentially, too. :)

Less popular as a Google search doesn't necessarily mean less popular, in some cases. For instance, I saw text messaging on your down trending list but that is obviously as popular as ever. People probably just don't feel the need to search it because it has become so ubiquitous.

That is true: this only measures current interest levels. A decrease in the current interest level of text messaging only means that people are not trying to find out about it as much.

However, why did it only start trending downwards in 2008? Text messaging was ubiquitous way before 2008.


Could it be the iPhone? Could it be the iPhone facebook app? Could it be BBM? (Or maybe I'm just reading too much into this...)

In regards to the mac screenshots, I found the key combination a little too much for something I tend to do a lot. On my Win7 laptop I have OneNote and the screen capture with <win>+S was great. Now that I have Skitch, I find it way better because of the key combo and annotation features.

On Windows "Print Screen" and "Alt + Print Screen" also take screenshots of your entire screen, and of your currently focused app, then stick it in the clipboard for pasting elsewhere.

For a steady decline, I got Christian music, flash, and Japanese dictionary stuff.


Similarly rising curve (although more steady) shows very different results. http://www.google.com/trends/correlate/search?e=id%3AMSXZzlt...

json, discogs, log in, landing page, haml, wordpress seo... I wonder why is "delete gmail" there.

Its surprising there are so many tube's.

The "comic book" link on Correlate is awesome. More sites should have documentation like this. http://www.google.com/trends/correlate/comic

Regarding the Google Flu trend, a Wired magazine article talked about how Google could identify the location of flu outbreaks before the CDC could solely because of search (http://www.wired.com/magazine/2010/06/ff_sergeys_search/all/...)

In epidemiology, this is known as syndromic surveillance, and it usually involves checking drugstores for purchases of cold medicines, doctor’s offices for diagnoses, and so forth. But because acquiring timely data can be difficult, syndromic surveillance has always worked better in theory than in practice. By looking at search queries, though, Google researchers were able to analyze data in near real time. Indeed, Flu Trends can point to a potential flu outbreak two weeks faster than the CDC’s conventional methods, with comparable accuracy.

Google did this when they first launched Chrome too - http://www.google.com/googlebooks/chrome/index.html

One (interesting?) thing to note is the Chrome comic was drawn by Scott McCloud, professional comic book artist extroidinare.

The Correlate comic was drawn by Manu Cornet, who's a Googler: https://plus.google.com/106154493978299996445/posts

This is fun. But apparently not "great fun" because according to this, great fun only happened in between 2005 and 2010.


Things are never as fun as they were back in the good old days. :)

From the whitepaper

"Google Correlate employs a novel approximate nearest neighbor (ANN) algorithm over millions of candidate queries in an online search tree to produce results similar to the batch-based approach employed by Google Flu Trends but in a fraction of a second."


Good, so i dont have to worry i m bringing google down / google is going to put a cap on this

Ok, that is way too addicting to visit during work hours.

I drew a standard bell-curve peaking in 2007 and then trailing off to zero into 2011 and everything was either Myspace searches or Washington Mutual credit searches.

Pretty nifty.

I did a steady growth curve followed by a steep drop off: The result is amazing: http://www.google.com/trends/correlate/search?e=id:lSkySxGeo...

If you like the drawing interaction itself and want to re-use it elsewhere, there's a self-contained version at http://dygraphs.com/tests/drawing.html which also features a zoom and eraser tool.

How is this so fast? I thought to compare charts like these you have to slide every chart over every offset to get the best match?

(I work on correlate)

The gist of why this is so fast is:

1. For the vast majority of queries, we only need to compare hashes of the time series, not the time series themselves.

2. We make no attempt to detect shifted correlations. The correlations you see are all synchronized in time.

They detail their algorithm in the "Methods" section at the end of their whitepaper: http://www.google.com/trends/correlate/whitepaper.pdf

1. Take historical end-of-day market data for a stock market index fund and graph it on Google Correlate.

2. Find out what terms closely track said index.

3. Set up an auto-trader that responds to changes in the search frequency for said terms.

4. Sit back and watch the money roll in.

Possibly, but apparently 5 years of USD to EUR exchange rate data most closely correlates to searches for the king tut exibit!


That's awesome! Bizzare correlation, but awesome.

How did you graph the data? Did you enter the time series data manually?

Er - no. Downloaded from oanda.com, uploaded to google correlate.

if only we could get better time resolution ;)

Correlate is great, but the hand-drawn year-by-year search is really not all that useful - wither you have something in mind and prime your graph, or you get the obvious things.

Monthly correlations, however, would be tremendously useful. Every industry has its own calendar, conference season, and economic 'tides', and understanding these is a key marketing task.

Google is beginning to do things that I think are going to kick current market-research strategies off their feet. They have most of our lives being fed through their systems; they know what's going on in the world better than anyone.

And if you visit the parent page (http://www.google.com/trends/correlate/), you can search for a term and see it's graph, along with strongly correlated queries.

This can be an interesting way to identify hidden yearly trends too.

Drawing a zig zag with peaks (more sawtooth than sine wave) at beginnings of the years yields some interesting results and some things I've never even heard of like "slapped cheek syndrome" or fifth disease.

Apparently people are also extremely curious about the carbs in a grapefruit/orange at around the same time...



I think you're seeing new years' diet resolutions.

Quite the opposite; the web is becoming so ubiquitous that no-one has to find out more about it anymore.

Inspired by this, I tried "the internets". I almost "wtf"ed out loud: http://www.google.com/trends/correlate/search?e=the+internet...

edit: and then laughed at the highest correlation with "the internet": http://www.google.com/trends/correlate/search?e=the+internet...

"Free Text Messaging" is an interesting 4 year period, though it looks like it's dead as a search term now. Is this because of Google Voice? Or did people give up on the idea?


Everybody has unlimited texting on their phones now so there's no need. There was a time when unlimited texting was rare and not all phones even had SMS, so it was popular to use various free web-to-SMS gateways if you were at a computer and your recipient may not be. I used AT&T's gateway in particular to message friends for a few years before I got an SMS phone. There were also shady ad-strewn "free texting" gateways, some acting as frontends to the official ones, some with a Turk behind the scenes, and even some that didn't work at all but just faked a texting interface to serve up ads.

I'd say a combination of Facebook chat, Google voice, and the announcement of Apple's new exclusive free-texting club. http://www.apple.com/ios/ios5/features.html#imessage

I drew a random hockey stick. It kind of matches the rise of dubstep pretty closely http://www.google.com/trends/correlate/search?e=id%3AOjXjy3O...

I got "youtube mp3 download" with a similar curve.

So it looks like 'Depression' is on the school schedule (see the spikes down for winter break, the heavy correlation with academic topics) https://www.google.com/trends/correlate/search?e=depression&...

I wonder how much is psychology research, and how much is side effects of school?

Some of the coincident correlations are really funny. https://www.google.com/trends/correlate/search?e=depression&...

It looks like OS X Lion had an unofficial mascot! (or the movie promotion timeline matched the lion hype buildup)

GIFs are coming back. I believe it is because the huge popularity of GIF flooded Tumblrs of teenagers: http://www.google.com/trends/correlate/search?e=id:4voTkH47-...

Also, it is nice to see summer related results: http://www.google.com/trends/correlate/search?e=id:Rhcg3luEC... Spider Vein, according to Wikipedia, may be caused by "environmental damage such as that caused by sun or cold exposure". And Asiatic Lily is a flower that grows on July.

0.9665 correlation between "iphone" and "cancel tmobile". Amazing.


I got this on my first draw. I was looking at things growing rapidly from 2009 on. What are the odds? http://i.imgur.com/yYL1D.png

Testing a yearly and incremental peak give me the "nc state basketball". I guess something like this could give me the Oscars or Superbowl changing the peak time: http://www.google.com/trends/correlate/search?e=id:nspW2uQjx...

Very cool but quite useless. I guess you might be able to spot emerging trends if you spent a bit of time on it though.

It's not useless. It has real world academic use cases where researchers need data which might be related. It's also documented in the comic book of this service. Also this is a service becoming real because of a real world problem.

It's useful yes but the usefulness is derived from the data not people drawing lines on a screen! I don't think the drawing a line on a screen allows for much more data discovery .

Yes sure. But it can be used by dataset input too.

"Windows Mobile Pocket PC" , "Palm Treo Software", "Blackberry forum" and "flash flv" all have the same curve :)


Hah. I did an exponential curve and sure enough there was "dropbox" at 0.9220 correlation.

My exponential curve beginning in 2010 yielded "dubstep download." I guess we know what the kids are listening to these days.

Apparently the acronym "OOMF" is hot right now: http://www.google.com/trends/correlate/search?e=id:lhfq3Utw3...

since I didn't know it is one of:


I post it so you don't have to google ;p

The comments here suggest that many folks don't realize you can correlate to a precise dataset, not just a hand-drawn trendline.

Click the link next to the search button. "Enter your own data" or "Edit this data".

Anyone got a good time series?

Different time period gives “Oops”



Git has apparently grown right alongside hackernews. http://www.google.com/trends/correlate/search?e=hacker+news&...

"Hot gf" is trolling a lot of these time series.


Makes me wonder if web developers finally switched to full SSL, or if browsers stopped showing the error…

I wish there was someone I could talk about this IRL.

Google Correlate as well as Trends are awesome. Why is google providing these data mining services for free to everyone? How are they making money from it?

This is seriously an awesome proof of technology, but what is a practical purpose for it?

Apart from just being able to search by drawing (cool concept in it of itself), the practical purpose would obviously be the reverse: looking at the graph of any search query. But obviously they will never release that :)

The main part of the correlate site lets you do this:


Google Trends?

Doesn't work on an iPad.

Well. That doesn't work on an iPad.

This would be infinitely more useful if it was backed by some actual data instead of those utterly useless search activity patterns. Why not combine a search like this with something like Wolfram Alpha?

What a waste...

edit: Seriously, wouldn't this be so much cooler if it gave us a new way to explore mathematical and scientific data?

If we could draw any curve and it would present us with matching constants, physical laws, mathematical proofs & theorems that have some correlation with the input data, chemical and biological data points, information from sociological studies, historical statistics...

Registration is open for Startup School 2019. Classes start July 22nd.

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact