Slashdot Mirror
What's Wacky with Google?
There are always going to be oddities with any big online service, but this one seems to be persisting. Join the discussion in trying to figure out a pattern. For maybe a week, Google has been returning zero results or "1-1 of about xxx,000" for common searches. One-word searches seem unaffected, but there are certain two-word combinations of common words like
candle truck
or
speaker bracelet.
Reversing the order can affect searches too:
motorcycle candles
vs.
candles motorcycle.
The strange thing is that usually the 1 or 2 results found are to commerce sites. Read the
Search Basics,
compare your notes to
GoogleWhack's,
have fun looking for patterns, but remember that Google always returns slightly different results for different IP numbers.
(Update: 13:56 GMT by J : When I first posted this story it said the problems have been occurring "for several weeks at least" -- but it seems to be more like one week.)
It still cannot do phrase searches:
"to be or not to be" produces a 20% error rate on the first page of hits.
Someone told me that this is OK, since Google is producing pages that are linked FROM pages containing "to be or not to be", instead of pages actually containing the phrase. What a cockamamy way to run a search engine. Altavista, a thing of the past, had its problems, but at least it could do phrase searches accurately.
I also keep getting searches where Google tells me that it could not be bothered to produce correct results, so it excluded certain words from the sentence, and I have to try again with a + in front of the words. Well, Google, I wanted those words in the first place, which is why I included them in the phrase.
Is 100% accurate matching results to a phrase search too much to ask for a search engine?
What possesses someone to try such weird random words in google. Thats the real trick...google wrote an engine to amuse the crazy users.
That's why you can't trust google for anything critical. You are at their mercy, and if they choose to do biased, or screwed up searches, you either don't know, or can't do anything about it...
I propose an opensource web based search engine... No more weirdness, no more screwups, no more censorship!
---
Programming is like sex... Make one mistake and support it the rest of your life.
Check out this - all 25 hits on the quoted words "candle truck" should be showing up in the non-quoted search ...
for a few weeks, when I do a search on google groups, it'll come back with the results just fine - but when I click on the View Thread on a result, it tells me it can't display the thread and gives me a link to view that individual message. Then once that message comes up, I click on View Thread on that message, and up pops the whole thread, like it should have before.
:)
Perhaps being on the top is getting to their CPU's
or maybe its the Google dance!
-dk
I realized the other day that although searching for 13 - 867 - 5309 causes google to go into calculator mode, searching for 123 - 867 - 5309 does not cause google to use calculator mode.
All sorts of odd things will both pull up an answer from google's calculator and also do a search - for example, searching for avogadros number or hbar.
So why do searches that might fit US telephone conventions not trigger calculator? Is it because some design decision makes it impossible to trigger both calculator and their phone lookup service. (Yes kids, google is a reverse phone directory, albeit with old data)
"q=site:www.google.com google" - (third result)
This is what I'm seeing...
http://www.sminkybang.com/google.png
By the way, for info on Google's purchase of the search engine Kaltix, check this controversial Register piece by Andrew Orlowski. It contains the highly suspect, matter-of-fact comment that "PageRank is now widely acknowledged to be broken," but if you take the PageRank speculation with a grain of salt it's an interesting read.
I think it may take a while for the stories to move through the cluster (there is a cluster right) because when I first saw the story the link "Read More" told me that I couldn't view this page unless I was a subscriber. I clicked it a couple more times and the same thing, then I reloaded the first page and I could get through. Probably some kind of concurrency bug involving subscription auth and story posts.
For any who are interested, Google.ca is behaving correctly. All search results listed (that I've tried so far) from googlewack.com are working properly and returning 1-1 of 1, or displaying as they should.
I wish I could compare to google.com, but for the past year or so, google.com automatically forwards all canadian IP's to google.ca
0110100100100000011000010110110100100000011000100
I know personally when i've been searching google of late for things like home improvement how to's and the like such as bathtub refinishing it is linking to TONS of commercial sites selling products and service but hardly any online howto's or guides. Granted I realize maybe there just isn't much content for these topics but google seems to be selling out more and more to commercial links. I've also notice this although not nearly as much in looking for other things more and more and some of the searches are for things listings etc which could not likely have a commercial equivalent or likely reason to be on a commercial page.
Does anybody else see the story change? I'm getting two different versions if I reload. One with the additional lines:
"The order of words matters also, with motorcycle candle revealing different results to candle motorcycle."
"Read the Search Basics, compare your notes to GoogleWhack's"
and one without.
Complete text of the two versions are:
"There are always going to be oddities with any big online service, but this one seems to be persisting. Join the discussion in trying to figure out a pattern. For several weeks at least, Google has been returning zero results or "1-1 of about xxx,000" for common searches. One-word searches seem unaffected, but certain two-word combinations of common words like candle truck or speaker bracelet are affected. The strange thing is that usually the 1 or 2 results found are to commerce sites. Have fun looking for patterns but remember that Google always returns slightly different results for different IP numbers."
and
"There are always going to be oddities with any big online service, but this one seems to be persisting. Join the discussion in trying to figure out a pattern. For several weeks at least, Google has been returning zero results or "1-1 of about xxx,000" for common searches. One-word searches seem unaffected, but there are certain two-word combinations of common words like candle truck or speaker bracelet. Reversing the order can affect searches too: motorcycle candles vs. candles motorcycle. The strange thing is that usually the 1 or 2 results found are to commerce sites. Read the Search Basics, compare your notes to GoogleWhack's, have fun looking for patterns, but remember that Google always returns slightly different results for different IP numbers."
Strange.
what if such OSS search engine is massively distributed?
Since by its nature search engine is not a transactional application, it can be effectively broken into thousands and thousands of semi independent pieces (just like real Google works now).
Anyone aware of Distributed Open Source Powered-by-people search engine project?
searching google for stone dog quote returns no results. Also try stone cat quote or changing the order of the words for weird results. Queries on alltheweb or altavista return numerous results, as expected. This has been reported in threads in alt.usage.english, rec.puzzles and (of all places) alt.fan.tolkien.
You should post your exact search, and what exactly you're searching for, if you want some help on this. The entire first page of the search I linked was (arguably) useful.
Oh, and one detail about your original post: The bit about which pages are shown is a little backwards. Google ranks pages with your search on it based on how many other pages out there link to the candidate page (regardless of their content, or at least without specific respect to whether or not those pages contain the search terms). I'm sure it's more complicated than that, but that's the short version.
So, if you're searching for "Xentax", your top results will be pages with Xentax, which are heavily linked to from other indexed pages. It's good for finding *popular* pages relating to a search, moreso than finding obscure webpages (by design).
Xentax
You shouldn't verb words.
For even more fun, use the following script to generate two random words:
/usr/share/dict/words` /usr/share/dict/words`
(watch for word wrap)
#!/bin/sh
#
dl=`wc -l
RND=`date '+%H%S%d%M'`
RND1=`date '+%y%S'`
RND=`expr $RND + $RND1`
bilge=`expr $RND + $RND + $RND + $RND + $RND + $RND`
dw1=`expr $RND % $dl`
dw2=`expr $bilge % $dl`
echo `sed -e ${dw1}p -e ${dw2}p -e d
So far, "pectoral undaunted", "adjudicates battlefield", "numerous quark" and "camouflaged todays" work as expected in google.
I've been getting an issue for weeks now where I'll do a newsgroup search, click on a match's "Show Thread" link and get an error that the thread isn't available. If you go back and try again, it works. Annoying, but not life threatening, at any rate...
... and the crazy users wrote scripts to use the Google engine!
(shameless self plug) Its surprising what sites can appear when querying Google. Try my site that queries Google with random words to find random webpages. Its quite powerful and a good timewaster.
google uses tons of DB entries to cross-index pages. I wonder if there's some simple hash-tables per page that it uses internally to speed things up that makes assumptions, and doesn't resolve collisions.
So you can search for one thing, and conceivably the checksum/hashes for each term match those of another page that has nothing to do with it, and it's returned as a relevant match by accident.
This might explain a lot of result sillyness.
Fuck Beta. Fuck Dice
Here's Googles somewhat hilarious cache of the Mamufilms.com page. The page includes links for everything from "Peter Paul and Mary mp3" to "preteen bra images". The text is vaguely reminiscent of actual gramatical English. Here's one sentence:
...Nothing interesting here. Just move along...
I made the same search and got 447 results.
o_0
Google reps is going to be on campus (Purdue University) today for a technical presentation. If there are any questions you want me to ask them reply to this thread. I'm 99% sure I'll be able to attend.
If you search for Candle Truck you get the single result. If you search for Candle Truck -basket, that excludes the one result and leaves you with... a working search with 80k results.
Yay for broken indexes
Here's a picture of this thingie.
I would argue you get one link to a page that does not contain the phrase - the "2Bee or Nottoobee". The other one I assume you're referring to is tobeornottobe.com, which I think is an absolutely sensible thing to offer when searching on "to be or not to be".
So, at minimum, I would accept that there is 90% accuracy in its top 10 results.
That said, I think I (and a lot of other people) disagree with you about what a search engine is. You compare it to a database query, which is sensible, seeing as a search engine is obviously going to depend on a database. However, I think the databse query is merely a process for answering what is, ultimately, not a database question but an ontological one.
When I type "to be or not to be" into Google, I am not merely querying a database for all pages that contain that phrase, and asking them to be listed according to a nebulous page rank. I am asking the far more subtle question "What page am I looking for right now? I'll give you a clue - 'to be or not to be'."
Thinking about the latter question instead of the former may clear up some of your problems.
Philip Sandifer's academic website
" Google doesn't do simplistic phrase matching. If it did, it'd be the same (and as useless) as altavista" Google does relevancy searches
No, it would be more useful by being as useful as Altavista. Altavista might have fewer results, but they are 100% accurate and thus relevant. Google's bad results are less relevant since I never asked for them. Non-matching results are not relevant unless they are specifically asked for.
"obeornottobe.com is relevent to a search for "to be or not to be"."
No, it is not, since I never wanted it. Sloppy logic is a flaw, not a strength. Sure, if I search on "Bill Clinton", you might similarly argue that "Al Gore" is "Relevant" and thus pollute the results with Gore results that never mention Clinton...
How aware are the search engines of each other? These returns are pages found searching in these search engine databases for themselves and others. Google and Yahoo rather casually mention the number of pages returned, prefaced with "... of about ..."; while AltaVista and Lycos are considerably more anal about reporting quantitative findings.
Google AltaVista Lycos Yahoo
Google 93,000,000 5,817,435 22,483,511 24,300,000
AltaVista 2,050,000 1,821,362 9,179,642 3,090,000
Lycos 18,500,000 2,309,191 11,215,263 6,950,000
Yahoo 95,300,000 10,284,666 55,680,102 38,400,000
e.g. Lycos found 22,483,511 pages mentioning Google, while only about half that many mention itself (Lycos). Perhaps this leads to poor search engine self-esteem issues.
Further exercises in pointless database introversion are left to the reader.
I spoke with a friend who helps maintain the google engine. She said that they were running into some problems with a "cleaning agent." Because of all the sites taking advantage of the word revelancy, there are useless sites that simply have a list of words or phrases. It's been posted before that there are many pages designed for GATOR/GAIN spreading or other spyware/adware. She quoted the percentage of junk pages being at 35% to 40%. The cleaning agent was supposed to run through its own searches and check for junk and keep a log.
She didn't say if the problem was that the cleaning agent was clogging searches or if any logged junk pages had been blocked. If so maybe the agent is flawed. In any case, they've stopped using it for the time being.
append -porn to your search to get all the motorcycle candles you want, except the pornographic ones.
2*31*37*263
Sure you know about audio speakers, but what about bracelets? And if you went to a crafts convention and heard a SPEAKER talking about BRACELETS, but didn't remeber their name, how would you find out.
These may be obscure searches because you don't use them, but google is still broken. As a programmer, you've got to take these warning signs seriously because these problems will creep into "common" searches and then people will definitely notice.
Think about security groups that look for vulnirabilities in software. They must use uncommon techniques to do this, and very frequently the results are suprising. Software like this (google) must be predictable with common and uncommon use, otherwise something is broken.
The counts have been broken for the last five weeks. A count for the word "the" produced fairly consistent results until then of about 3.4 billion. Then it shifted five weeks ago to 5.2 billion. Lately it has been under 2 billion. Now it's just over 2 billion.
Webmasters who have various directories and know exactly how many pages are in each directory, began noticing five weeks ago that Google was reporting approximately twice the number of pages in each directory than have ever existed in that directory. Prior to five weeks ago, Google used to be fairly close to the actual number (assuming that you get a full crawl).
GoogleWatch speculates on the reason why Google has been behaving strangely ever since it stopped doing the traditional deep crawl once per month. The last standard deep crawl was in April but it wasn't used -- Google threw out this data (by their own admission) and reverted to earlier data. The speculative piece was written last June.
Since it was written, Google has started showing "supplemental results" on many searches. It looks like they are running a parallel index. Why would they do this? All the problems Google has been having, along with the supplemental index, seem to support GoogleWatch's theory.
Duplicating search terms has an interesting result:
candle truck
1-1 of about 101,000
candle candle truck truck
1-1 of about 82,200
candle candle candle truck truck truck
1-1 of about 73,700
candle candle candle candle truck truck truck truck
1-1 of about 68,600
Another interesting one is
candle candle truck
1-2 of about 89,200
Who are you kidding? There're probably a dozen google-gurus laughing at this thread as we speak.
"Hey, guys, you know that bit of code we wrote to screw up the google whackers? Slashdot finally took notice!"