|(last 3 letters of a 6 letter name) |
Based on this, I'd be reluctant to generalize and say they're matching abbreviations, but I haven't seen the particular example you've seen.
I think the way it's working is that if you type in enough of word or phrase so that it's statistically very likely that you're looking for a particular word or phrase, Suggest will fill in the rest.
Usually, as you've noted, it's the beginning of a word... but sometimes it's the end of a word or sometimes it's the middle of a word that will trigger the suggestion. I think it's simply statistical pattern matching.
Eg, typing in aho will bring up "yahoo", whereas typing in hoo will not.
I noticed something like this just today when I missed typing two internal letters in a company name. Yes, very fat fingered typing on my part, but Google's suggestions stuck out at me as something I'd never noticed before.
|I'd be reluctant to generalize |
Yes, I should have known better than to try that around here :)
OK let me rephrase "Google suggest apparently magically matching word fragments with the word I am seeking"
Thanks for the yahoo tip Robert Charlton - that is an excellent example of what I was trying to communicate.
Google's computing muscle never ceases to amaze me. The algorithm for this must be incredible
read next input letter, build ngram, search suggest table for ngram, if ngram = fragment of significant search term then suggest search term
Now I'd like to know how they decide that "aho" might be a search for Yahoo. I gather there'd be some 'significance' assigned to particular search terms.
Thanks for your replies fellows.
Just tried a couple and it seems to be on specific strings only. If I owned one of the sites legitimately using the term "Ahoo" I would be rather upset at Yahoo getting the top three results in the SERPS.
Look at the actual SERPS for "ahoo": after Yahoo comes an individuals Facebook page,an Iranian magazine, a Linkedin page, and a discussion (on a forum) abot Belly dance lyrics.
There are going to be FAR more people searching for yahoo who mistyped, than who actually want any of the possibilities for "ahoo".
If you try Hoo, they first show a few results for Hoo (such as Sutton Hoo), and then Yahoo related results.
What they are doing looks right to me.
I think there is an important distinction to be made here:
Are we talking about the Autosuggest when typing into the search box or the suggestions we see in the SERPs?
As some are talking about the SERPs I'd say that we are experiencing "Autoforce" on some terms, going beyond suggestions and actually favouring terms that are not what are asked for (that may be a good thing for searches that are typed incorrectly, but it's very annoying if you have taken the care to type correctly and see irrelevant results rather than a "did you mean" suggestion). In fact it's likely that the number of people clicking on the "did you mean" links (or other similar autosuggest features) are influencing what is autoforced and what is autosuggested. It's statistically correct for Google to show corrected results if 95% of people that type a give query go on to correct their query in some way, but it's still annoying for 5%.
Maybe there should be a setting that allows people to set whether they want to disable autoforce, or even have a few options as to how results are shown.
As for the autosuggest in the search box, that should always have the option to disable it - but at least you have the option to keep on typing if you don't see what you want. A big problem with autosuggest in the search box is that it guides people to searches that are often close to what they want, but not what they would have typed, so some of the query is lost when you have close suggestions - this may have unintended consequences for advertisers (good or bad).
As a wider consequence of autosuggest, Google is able to show results that it has more information on, and hence Google appears smarter. There are serious problems with the way that possibly good matches to a query are discarded in current search engines (because the page lacks a particular word), the way current search engines get around what is a real problem is to index huge numbers of pages and hope they have enough that carry all of the terms you are looking for OR they simply try to suggest that you search for something they have more info on, this can often be seen when you search for a phrase that you have appended rarely occuring word to (the search engine recognises that the rare word is knocking out many possible matches and suggests you drop it - although this behavior has been reduced of late in Google).
I suppose the major problem is that search engines have to cater for the reality that people do get their queries wrong, either through typing errors, inability to spell or through poor search ability (all those happen very often - and that "search laziness" is being reinforced by the fact that search engines deal with most issues quite well - why lear to do something right if it can be corrected 95% of the time?). Search is difficult enough, add in a layer of complexity about whether the search that has been sent is actually what was meant and you are in a real minefield.
Those of us that do know how to search, and spell and take care to check their searches should be given more respect by the search engines, meaning they should answer the question we ask - not the question that they think we have asked, but we are in the minority and we served our purpose years ago when we spread the word about a great new search engine. Are we now an inconvenience to Google? It seems so to me.
Why is it that I keep reading 'google suggests getting smarter'?
in a blog post they've stated that google suggest can work based on your personalized web history. I imagine that some of these examples may have come based on your past searching/misspellings/intent. If you've typed google in a thousand times, and then type in gle, I think that you should get a different suggestion than a typical web surfer.
trying all sorts of gle <insert product> queries does not give me any suggestions whatsoever.
there's options to turn off suggest based on personalized web history, and you can also log out of your google account to see what g users would typically see.
Good observations, rohitj - people should definitely keep personalized factors in mind.
My observations last week were all done logged out. I rarely search while logged in, and even then I've turned off history.
Some machine intelligence about near-match word fragments seems to have taken a nice leap upwards, even for non-personalized searches.
ohh those suggestion still gets on my nerves, I always delete cookies at the end of day, then I have to remove suggestions everyday or use #*$!.org even when you have changed settings, when you change to another country you have to change settings again.
If it bothers you so much, then you have a few options:
1.) If you use chrome, then you browse in incognito mode which will make it so that they don't recognize your google account.
2.) Use one browser exclusively for searching and never log in to your google account from it.
3.) Disable suggest all together. I wasn't aware that your settings vary by country. That's really interesting, though if that really is the case then I could see that as a minor annoyance if you are toggling between numerous google country-specific search sites.
Personally I think suggest is a great product and millions of people have reaffirmed that. Stats have shown that a lot of people do repetitive searching. In fact it's a staggering amount, like > 50 %. So this should help you spend LESS time thinking of queries and more time getting you the results you need.
google suggest has always been a favorite product for me as it helps immensely in my job. However, I agree what inbound suggested but i doubt if that is going to happen in the near future...
This is an extension of the "best guess" search suggestions that appear under searches, notice the list changes as you type each letter.
To add to my ealrier question I think there must be significance applied to ngrams.
These significant ngrams are the significant fragments of significant search phrases. (!)
So Yahoo is a significant search term and to use Robert_Charlton's example
- aho will trigger Yahoo
- hoo will not trigger the Yahoo
- yho will trigger Yahoo
So my question stands:
How does Google decide which combination of letters should trigger a search term, especially if it is a mistaken fragment?
Google has published several patents and applications in this area that can give us an idea. For instance:
- Anticipated query generation and processing in a search engine [appft1.uspto.gov] and
- Predictive information retrieval [appft1.uspto.gov]
Thanks for the links oh Brainiac tedster </fanboy>
|... the portion of the search query received ... may be compared against the entries of the dictionary ... As a result of comparing the partial query against the dictionary, entries in the dictionary matching the partial search query are then returned ... Various criteria may be used for identifying matching entries when a partial query is compared against the dictionary ... |
And they have multiple dictionaries? Wow.
This pops out (I don't read patents much and I assume it is a standard patent phrase)
|Those of ordinary skill in the art |
Blimey - this sort of engineering is considered ordinary?