Google Recently Made A Silent Shift To A New Search Algorithm, “Hummingbird”
techcrunch.com
techcrunch.com
It's always worked reasonably well and saves me from trying to come up with a search.
At the same time I've noticed that coming up with a sequence of search terms has been working worse and worse in google over the last couple years. I frequently get results for whatever google thinks I was searching for, especially if my original search terms resulted in very few results or no results, I'll just get a result page anyway except it's almost never helpful.
Perhaps this is an attempt to make Google more Star-Trek/Watson-like, and it's great for those use-cases. But for the other cases, like looking up specific serial numbers or whatever, it's a mess.
Who said, "To hair is human?" exact: Who said, "To hair is human?"
Thanks for pointing that out.
"heir" is a much closer match for "hair" in terms of Levenshtein distance[1] (which indicates a typing error) and soundex[2] (misheard). But clearly "to err is human" should be offered as a possibility too.
@nimble: what about when you intentionally search for a misspelling?
Usually you can click "search for {what I actually typed}" but it's still annoying. I liked it when +'s and double-quotes consistently did the same thing every time.
Maybe there is a setting that does this still? I've tried playing with a few settings but have had mixed results / consistency...
However, I just tried a search for "anagram +python" and got 67 results vs 139,000 without the plus-sign. Does that mean they quietly restored the operator?
I always use it to filter out commercial or store results when a query triggers some over-zealous SEO and throws up pages of junk. A quick "-buy" or "-store" usually cleans the results up.
Google results are getting worse and worse IMO and there don't appear to be any tools left in the available query language to rectify it.
I honestly suspect Google is A/B testing search results somehow.
They're optimising for normal people who don't formulate query strings in Google, but they ask questions or just search.
Apparently the double-quote still works. Well, that's good.
But it seems inevitable that as google targets what most people mean, they will target less what niche people mean (us). They could do both, with explicit operators etc, but would double their indices, and not worth it (for them). :( At least it creates opportunity for the google-usurper.
Then there's "file:" and a file extension name to find searches of that filetype. (example: "AP biology file:ppt" returns all ap biology related powerpoints.
Next there's "site:" and a website. (Example: "site:reddit.com fedoras" searches for fedoras on reddit, a very common thing. Useful if a site's search engine sucks.)
I often simply ignore site search boxes in favor of using the "site:" operator.
[1]https://www.google.com/cse/
[2]http://www.google.com/enterprise/search/products/gsa.html
Searching for a few random queries with either "file:pdf" or "filetype:pdf" shows that both return pdfs, but the results that I get are different. Not sure what to make of that or which is better.
[1] http://www.googleguide.com/using_advanced_operators.html
"filetype:pdf" seems to be the only command which reliably returns just .pdfs.
I'm in the UK, your mileage may vary.
It's worked out really well. That plus the !g operator in ddg means I get the best of both worlds, I'd suggest you try it for a day or so and see how it goes.
This is an interesting area where I found myself playing catch-up to the less technical people in my life. For a long time I saw them typing questions into google and getting bad results and my recommendation to use keywords never really stuck, probably because they didn't have the same mental model of how searching worked that I did.
Then along came a few sites that targeted question-askers and sometimes if you asked a question you'd find someone else asking the same question along with some answers of high variable quality. I didn't discover this myself, because I never typed questions into google -- I had to see observe the less technical people getting better results than me occasionally to pick up on this.
Still, they got worse results most of the time, and while it was a new tool in my arsenal, it wasn't usually what I was looking for and I used it sparingly.
Then when the first iPhone with Siri came out, I bought it for a family member and demo'd it. I did all the stuff the commercial was doing to make it seem like magic.
So my family member takes it and starts talking to it like it's a human. And I instantly regret what I've done, because I knew some keywords it would pick up on to look like magic, but I've made it look like you don't need to know the keywords, you just need to talk to it. So there they are, talking to it like a human, and I'm expecting it to fail. But instead, when they say "call my sister" it replies with a prompt asking who the sister is. I'm sure Siri is still a large number of special cases, but it's large enough that thinking of it that was failed me.
Similarly, I noticed one day that people asking google questions were getting better results than me sometimes even when they weren't looking for other people asking/answering the same question. So I've started asking google questions more often, but again, it required me to observe a less technical person doing so.
That's a really fantastic observation. I suspect that lots of these changes are meant for those kinds of people (who happen to make up the majority of the world) and not for tech folks at all.
But yeah, I notice that if I ask a question, quite often the results I get are on question asking sites like Stack Overflow -- it turns out I usually get very good results that way too ;)
For example say my question is: "how to use rails with devise and omniauth", I break it down to the group in which I think nets the most results should be the first keyword and so on.
1. rails 2. devise 3. omniauth
rails+devise+omniauth: https://www.google.com/search?q=rails+devise+omniauth
rails+omniauth+devise: https://www.google.com/search?q=rails+omniauth+devise
omniauth+devise+rails: https://www.google.com/search?q=omniauth+devise+rails
how+to+use+rails+with+devise+and+omniauth: https://www.google.com/search?q=how+to+use+rails+with+devise...
The top results are normally similar. With exception to the "how+to+use+rails+with+devise+and+omniauth" search, which is rather different.
From my standpoint the "rails+devise+omniauth" yields the best results as the source of truth is closer to the github documentation "OmniAuth: Overview · plataformatec/devise Wiki · GitHub", over third party information "#235 Devise and OmniAuth (revised) - RailsCasts".
Using "how+to+use+rails+with+devise+and+omniauth" gets rather off topic after the 3 or so results.
Another tactic I use is searching like a command line. For instance: "wiki list breaking bad", or "imdb iron man". This don't work with smaller properties, so "site:example.com" is another awesome tool.
Just my two cents.
I would love an advanced search mode with some simple boolean logic, even if it took several seconds to return the result, for those times when you really need to sift through a lot of garbage to find the results you really want.
What is the point of Google returning results in 0.000000001 seconds if they aren't the results I wanted and I have very little means of refining the search. The moment a new search engine comes with this functionality and a decent indexed base, it's bye-bye Google for me.
That being said machine learning is far from perfect. Allowing for user correction is still an immense must.
Let's hope basic search operators like quotes don't go away anytime soon.
I wish the 1998 Google page that made the front page on HN was real (with updated index of course).
It is a boatload of extra typing, but the "allintext:" operator is your friend for those kinds of searches. I wish there was a URL modifier that could force google to assume allintext: on the search - that way I could have two different google search engines in firefox, one for literal and one for more "semiotic" searches. Maybe there is and I just don't know it?
http://www.googleguide.com/advanced_operators_reference.html...
[1] http://support.mozilla.org/en-US/kb/how-search-from-address-...
EDIT: Looks like Chrome works in a similar manner: https://support.google.com/chrome/answer/95653
Edit:
* Links to the websites, I would say almost identical
* Same number of the pages are indexed by google, around ~5 million
* Domains are almost same, no keyword difference
* They both have same pagerank
* Domains are registered together
* Sites are hosted on different ips
* Total traffic sites get is around ~40k/day unique
* By this change, total unique increased by %10LSH: http://en.wikipedia.org/wiki/Locality-sensitive_hashing
TFA mentions this update happened weeks ago. How are you able to pin this down to the hour and correlate your traffic changes to the algo change if we don't know when the algo change happened exactly?
Where is disneyland - Shows a map and has "Get Directions" https://www.google.com/#q=where+is+disneyland
How big is disneyland - 160 acres https://www.google.com/#q=how+big+is+disneyland
Not sure if it's new.
"What is PI?" shows me a calculator with the result.
It makes me wonder if Google has become so "good enough" that more and more engineering effort will be spent for smaller and smaller returns.
How do you know it was the "biggest overhaul"? Google is known to use every moment to hype itself.
Also Google has been doing these changes since 2011 almost monthly. Now sites that sell things or make money have almost given up on "free" traffic from Google, they know it's pay to play (outside Product Search it's not openly pay-to-play but you get the message after a few 70% traffic reducing updates).
Unfortunately the opposite example can be found in the same product, Gmail; when they visibly and ostensibly changed the UI, it was immediately noticed, and widely hated.
It isn't that the special effect doesn't have an impact, but that you don't specifically notice the effect because it blends seamlessly into the rest of the film.
Okay, I guess it depends what you mean by "notice." I certainly was aware that much of Titanic was shot in front of a green screen. No matter how photorealistic special effects become, it's often fairly obvious what's real and what's not.
* There are a ton more people actively researching ways to reverse-engineer and game the system.
* There are a ton more sites and more content getting generated, including bot-generated ones which are of dubious quality.
Dictated queries are going to come as natural language, not query syntax.
I dread the day I'll have to look for some satiric blog called something like 'how to not' but I won't be sure enough about the exact terms to allow me to search verbatim.
It feels like there's an expanding dark space in the shadow of bright neons.
I wish I could see how it affects their AdSense revenue. (a lot of 'answer' sites are just AdSense landing pages)
At what point does it take all the load off the Wikipedia servers if it just serves up the answer on the page :-)
There are many things that I think should be google-time-worthy:
- mail notifier as a first class citizen, a tiny dot would be a time saver
- many of their urls are 256+ chars in size ... , I'd love parseable urls that fit in a 1024 width browser bar
- no more link hijack on the front page, they surely know how to do it without rewriting the link on click ... or at least propose a quick way to select the original url shown below (a web semantic-unit/best-practice since long)
https://www.google.com/search?q=How+has+Google+changed+its+s...
During the past decade, Wikipedia emerged as the primary source of high quality organized digital knowledge. This, together with the open data policy practiced at may governmental agencies, supplied search engines with quantitative data easy to process and understand.
For the next few years, this looks like the biggest change in search engine technology to expect. We are going to switch from mathematically precise queries and semantic processing of the search corpus to a semantic processing of the queries and a reliance on more quantitative data sources.
What's wrong with answering questions and offering everything under the sun? Because another site might have better answers and offering but Google's MBAs will make sure that page is buried down below--unless they advertise.
"Best for the users" has been replaced with "best for this quarter's earnings," and it's just a matter of time before it catches up with Google. Google criticism was barely existent just 2-3 years ago, now it's widespread. What else will they do to prop up their earnings??? On what can you trust them?
How about you tell me line by line where I'm wrong? Beware that I do this for a living, I help manage quite a few small business sites and know this first hand. I also follow Google's stock and earnings reports. You should read Patrick Pichette's comments about their low CPC and how ad clicks increased by double digits. It's obvious that Search will hand more penalties this quarter to more sites, hoping they advertise and help increase CPC along the way.
I'm afraid that it would still not convince you otherwise.
Name them. And how much money Google is investing in them. I could launch a project that costs me $5. All these "projects" are tolerated because Google is increasing earnings by double++ digits each quarter, albeit by destroying every bit of their credibility. The minute the over-monetization growth stops, Larry Page will see Google's stock drop by $200+ a share and he'll face a mutiny within Google and the press will call Google a has been. You bet he cares, he's just a good liar.
Obviously you don't expect a secret email or memo, we deduce from their actions and results.
Obviously nobody is expecting a leaked memo or something, but an explanation of which actions and what results you are talking about would be nice, since you are the one making claims.
At any given point there are 10 sites that are listed in the search results and millions of others that need to advertise to get to the first page. Those millions will always think that Google is out to get them and that some conspiracy is involved.
You do not know much about search engines. Maybe a Googler that optimizes Search to get more ad clicks can come in and explain it better, but if the "10 results" are buried deep, barely anyone will see /click them.
Then it should be pretty easy for you to point out some factual account of an action or property of Google that backs up your claims. Feel free to start with any sentence you wish.