Home > Contributors > Blog

Blog

Search Engine suggestions welcome

Permalink Comments (40)17 December 2007 at 11:28 by Pippa Greig
Posted under News

The Alamy search engine has the ability to interpret keywords in a search and retrieve images that contain variations of the same word. Occasionally this system makes incorrect assumptions about which words to associate with the original search term. It was recently pointed out on the blog, for example, that searches for 'executive' returns the same results as 'execute'. Rather than execute our executives, we'd like to invite you to submit any other examples that you feel are generating irrelevant results.

To report examples to us please email Member Services with the subject line "Search Engine suggestion" and state which words have been incorrectly associated.

Please note that although we will carefully consider all of your suggestions, we cannot guarantee that every suggestion will be implemented.

Top

Add your own commentComments (40)

  1. 17 December 2007 at 16:07 Jeff Greenberg

    celebrate
    celebrity

  2. 17 December 2007 at 16:11 Itani

    (orient express )= (oriental expression)

  3. 17 December 2007 at 16:30 Vidra

    Laden
    Bush

  4. 17 December 2007 at 23:25 Keith Erskine

    Search was: Keith HARING
    Result HARE Krishna and copyright KEITH Erskine

  5. 18 December 2007 at 08:04 Rainer Raffalski

    organic = organization

  6. 18 December 2007 at 09:55 Duh!

    Read the post properly guys. It very clearly asks you to EMAIL the examples. They must despair over some of the stuff they read here.

  7. 19 December 2007 at 21:50 Think!

    Nowhere does it say not to post here, too. Duuuuhhhhh.

  8. 19 December 2007 at 22:16 Kevin bailey

    Create the ability to string two words together (maybe more) so they read as one in a search by using a special character around the specific words to create a hard containing break; i.e. quote symbols “keywords here”.

    Here's the relevance: type in the word acorn and not only will you pull up acorn pictures, but whales too. Why whales? Because the whales have "acorn barnacles" on them. Better examples can be found, but that was a recent one I came across. True, one can fuddle around in advanced search and trim the hits but why the bother.

    This will help buyers zero in on relevant images faster by eliminating unwanted hits, and also speed up the process of finding images.

    It will help contributors CTR score by insuring their "whale image" isn't going to get a hit when someone is searching for an acorn seed instead.

    I've seen some of my own images turn up in hits that weren't relevant to my intent, but likewise the keywords couldn't be edited to solve the problem either because in singular form they were entirely relevant.

  9. 19 December 2007 at 23:27 Thunk!

    Nowhere does it say not to tell everyone your PIN number, either. Double duuuuuhhhhhhhh!!

  10. 20 December 2007 at 08:08 MGrebler

    Kevin: This is already in preparation:
    http://www.alamy.com/contributors/stock-photography-syntax.asp

  11. 20 December 2007 at 15:16 Kevin bailey

    Appreciate the link. Very interesting, exactly as I was describing, but much deeper. Excellent.

    These further comments may be on the site somewhere that I haven't seen, but here goes anyway:

    –Other recommendations would be to provide a button for *illustration, *photo.

    –Consider moving all orientation buttons to the main page from advanced search since these buttons would be used frequently in a search. Advanced search being used strictly for refining key word combinations. This is a subtle quibble but one that would be viewed favorably by the buyer I believe because these are fast features to achieve relevance in hits.

    –Put a slot in search results to allow the buyer to key in and skip to whatever page they wish in search results. Important when a search brings up a large number of hits and the buyer can split duties of a search to several people each starting at separate page number. Also good to pick up when one can't peruse a search in one sitting.

    –Continue with the non edit policy; however consider an edit photographer policy that would review select photographer portfolios that show up on the radar after so many years without a single sale. Consideration would need to be given to the number of images that photographer has for sale, obviously one can have only a few high quality images for sale that don't sell because the market for them is slim, however if a photographer has a thousand images for sale and hasn't sold anything in say 5 years then a judgement can be made that the site quality and relevance will benefit with the exclusion of that body of work.

    As Alamy continues to grow its body of work you must at some point put in other measure that will help control quality so the buyer continues to have a rewarding experience in search. The CTR is a fantastic step, however consider other measures such as an occasional internal portfolio review which wouldn't require cost, manpower and judgement the way individual editing of images would require.

    –K


  12. 23 December 2007 at 20:13 Jacques Jangoux

    What about an optional random search? For the researchers who would have the patience to use it, it may bring up some pictures that are never seen. If a photographer has a bottom AlamyRank but a few great pictures, these pictures would get a chance to be seen. I will add that pictures that are clicked in a random search should jump up in searches. Another agency (that edit pictures, so the junk is left out) uses random searches. And I always wondered why Alamy uses contributor ranking, but apparently not picture ranking, which would bring the best-searched and the best-selling pictures on top.

  13. 03 January 2008 at 09:38 Shaun finch

    search for Bude (town in cornwall) brings up results for BUD as in plants.

  14. 04 January 2008 at 02:15 Bert Klassen

    stock = stocking

  15. 09 January 2008 at 03:23 Paul collis

    Why does Alamy want us to send in examples of stemmed words when they say they've already solved the problem?
    I thought that the new syntax recommendations found on the
    'Manage Images' page (> Additional annotation options > Exact match / Do not stem) would solve these problems.
    So I'm undertaking the laborious task of re-keywording 2,000 images. But as of today, those looking for a CARP will still be shown my images of a sundial with the inscription "^carpé ^diem".
    Those looking for PEBBLES will be shown images of cars at the "Pebble Beach Concourse D'Elegance".
    My question, Alamy, is: is there any real point in doing all this work? If so, when will these "-----" and ^----- annotations come into effect?
    Please say soon.
    Thanks.

  16. 09 January 2008 at 23:04 BH

    Paul,

    As I understand it, the annotations can stop your words stemming into other words - they can't stop other words stemming into yours.

    For example, a recent search for the word protest returned my image of a protestant church. If it was the other way round, I could ^protest - i.e. do not stem protest - but I can't prevent the search engine stemming protest to protestant as the word protest is not in my keywords.

    The exact match syntax ("") is designed to deal with phrases, not individual words.

    Also, I would suspect it would be better for Alamy to deal with some anomalies specifically, rather than wait for us all to change our keywords.

    That's the way I see it, but I'm willing to be corrected!

    Cheers.

  17. 09 January 2008 at 23:57 Paul ^Collis

    Thanks ^BH,
    but perhaps a ^person who could ^respond most ^authoritatively to our ^queries would be ^somebody from Alamy who ^actually "^knows the ^score".
    ^Hopefully we'll get an answer ^before all the "art ^buyers" and "photo ^researchers" get ^too ^tired to ^wade ^through "^11 ^million" ^images and decide to look ^elsewhere.

  18. 10 January 2008 at 22:22 Olga D. van de Veer

    A search for white van gave all my pictures with the keyword white because my last name is "van de Veer". I don't think the buyer was looking for that ;)

    Olga

  19. 10 January 2008 at 23:16 Richard Baker

    Compassion

    Compass!

  20. 15 January 2008 at 12:24 Rudi Sebastian

    As i am often searching images as afreelance cd, i can say that i very often know better what i do not want rather than what i want. For example when i am searching for images from venice, italy. I might know that i want something special, some not so well knwn locations and that i do not want the marcus square. Or i am looking for a great flower and the onlx thing i know is that i do not want tulips. Conclusion many problems could be solved, if there were a good way to exclude search results rather than to include. In daily business you often know better what you do not want. Even if the engine can do this, it should be much more prominent.

  21. 15 January 2008 at 12:31 John Gaffen

    Forgive me for mentioning a competitors website, but Alamy should perhaps take a look at how the Photoshelter search terms are entered. It allows contributors to closely control how each individual key word is defined, contextualized and used. For instance, choices of noun, verb, etc, choices of meaning, where a word can be used in different contexts or meanings e.g face, as in body part, or face as in rock face, or face as in be opposite to, they have a built in thesaurus which the photographer is made to use. This would help reduce stemming problems, stop irrelevant pictures appearing in searches and would pevent negative effects on photographers individual Alamy rank.

  22. 16 January 2008 at 06:47 Morten svenningsen

    (Not about keywords, still related to the search engine though:)

    Time to market: 24 hours!

    Nice improvement. It’ll open up the possibility for Alamy to license news photos as well. At least for second cycle use. Any chance that Alamy can make a little check box in the ‘Advanced search’ that would restrict search results to photos shot (not just uploaded, but actually shot) within the last 7-10 days? A “This week in photos”-sort of thing.

    Or generally make it possible to set start and stop date for search results?

    It’ll never be Reuters who have more strict policies about what a photographer can and can’t do, I know. Just wondered how it could turn out...

    Morten

  23. 16 January 2008 at 22:25 Jacquues jangoux

    To Rudi Sebstian (post 20)

    "if there were a good way to exclude search results rather than to include"

    Well, from the point of view of the picture researcher, there is the Advanced Search which allows both to exclude and to include.

  24. 25 January 2008 at 15:31 Shaun Finch

    Search for the town Bude brings up results for bud.

  25. 25 January 2008 at 17:27 Bert Klassen

    Community = Communicate

  26. 29 January 2008 at 10:22 Tony lilley

    I have found that when checking the actual pictures Zoomed against the number shown in the table, that the number in the table is less than the number of pictures displayed, when I actually look at them. Will this give me a lower CTR than I should have?
    Hope this all makes sense.
    Best wishes, Tony Lilley.

  27. 06 February 2008 at 18:07 John Gaffen

    a recent serch reported on Alamy Measures; Polarize brought up my images of Polar Bears!

    e.g Polarize = Polar Bear

  28. 11 February 2008 at 15:27 Andrew Rubtsov

    "boxing" found "boxes"

  29. 11 February 2008 at 15:39 Leslie garland

    Whereas encouraging contributors to be a bit more accurate with their keywording is obviously not a bad idea, I would suggest that much more would be gained, far more quickly, less painfully and much more effectively, by encouraging researchers to use more words in their searches.

    Inspection of my own AlamyMeasures (Beta) results indicates that by far and away the vast majority of searches are made using only two, yes two search words! Back in pre-online picture library days, picture researchers used at least a sentence, if not a paragraph, to describe each picture requirement. So there is not the faintest hope that they will be able to locate their requirement accurately with just two words. So it is no wonder that they are pulling up, what must be to them, thousands of irrelevant images, and no wonder contributors' CTRs are taking such a battering, even though their keywords are 100% relevant to their picture!

    Some simple real examples:-
    1) Using "newcastle" as a search word pulls up 4,825 images.
    Yes, I have just said the vast majority of searches use just two search words, however both this example 1) (using just the one word) and 2) (below) have appeared on my Alamy Measures (Beta). THESE ARE NOT MADE UP!
    2) Using "newcastle uk" calls up 2,504 images.
    3) However adding an extra word, let us choose "newcastle theatre uk" immediately brings the number of images down to 61.
    4) Then, actually using a more sensible number of words, "newcastle theatre royal uk" reduces the search to 24 images
    5) And finally, refining this to "newcastle theatre royal uk night" finally brings the search down to just 8 images.

    As example 1) did occur, I will not ignore it. This search will result in the researcher having to look through something like 193 screens of images, at 25 per screen, in order for them to find "Newcastle's Theatre Royal at night". OK, that should be properly regarded as "their own silly fault" and should not concern either Alamy or us contributors. Unfortunately, Alamy think otherwise, and 4,817 images will be classified as "views" by AlamyRank and the various contributors' CTRs penalised as a result. With example 2), it is pretty obvious that the researcher is going to have two and a half thousand pictures to go through to find pictures of "Newcastle's Theatre Royal at night", and in this case Alamy's ranking system will penalise 2,496 images.

    The philosophy behind AlamyRank assumes that the researcher requests exactly what they want and that anything pulled up by the search and then not zoomed must be "badly keyworded". Instead of just ignoring these images in any calculation of CTR (zooms / views as a %), AlamyRank includes them! So in the example above, all those 4,817 and 2,496 images will impact negatively on their contributors' CTRs, even though those images were correctly keyworded "newcastle uk", simply because the researcher was not specific about what they were looking for! And of course, there is absolutely nothing that the contributors can do to reedit these images to prevent this happening again, because they have no idea that the researcher was actually looking for a shot of the Theatre Royal at night, or for anything else for that matter, because it was not specified, and even if the researcher does subsequently refine their search to include extra words, the new, refined searches will not overwrite the previous vague ones, whose negatively impacting CTR will still go towards calculating the overall CTR for that Contributor!

    This really is a flawed system. However, making no further comment on the workings of AlamyRank, it ought to be patently obvious that encouraging researchers to use more words is going to prove far more beneficial to the "search and find process" than getting contributors to try and reedit their work.

    To this end, why not revise the search window? Perhaps break it into three panes (this might only be presentational, the actual window could still be a single window), with the first pane titled up as, say, "Subject", the second titled up as, "description / qualification of subject" and the third as, "geographic location / city, country, etc." The researcher would be prompted to fill in what they know about what they are searching for. Thus, with any luck, most searches might at least start off by using a minimum of 3 words, which has got to be an improvement on the current 2! And hopefully might use a more realistic 4 or more.

    Were this the case, this would have the double advantage of dramatically tightening their search and at the same time resulting in less totally unwarranted damage (in that this has nothing to do with bad keywording) to contributors' CTRs.

    So how about it Alamy? Think about revising the search window to encourage researchers to use a sensible, realistic number of search words for their searches, instead of the stupidity we have at the moment.

  30. 11 February 2008 at 15:50 Leslie garland

    Annotation options.

    Yes something of this type was definitely required. However, theory and practice have a habit of not being one and the same.

    Proximity; works well for the examples given, but what about "big large old ancient medieval stone building"? If proximity works effectively, it is not going to be very helpful here. Indeed, it effectiveness could well be detrimental in these keywords, or similar combinations, operating efficiently. E.g./ the words Big and Large could and need to relate to any of the other words equally, and not unequally depending on their proximity.

    Exact match; "Golden Gate Bridge" reads well and sounds logical, but in practice one will also have to have Golden Gate either as "Golden Gate" or as [Golden Gate], and one will also have to have Bridge as well. So the entire string will read "Golden Gate Bridge" "Golden Gate" Bridge, which I expect is too much for the Essential Keywords box.

    But try instead:- Lord Admiral Horatio Nelson, and it gets more complicated. If one is using exact matches, one ends up with permutations:- "Lord Admiral Horatio Nelson", "Lord Horatio Nelson", "Lord Nelson", "Admiral Horatio Nelson", "Admiral Nelson", "Horatio Nelson" and of course, finally Nelson! There is no way that this lot will squeeze into the Essential Keywords Box. Bearing in mind that all combinations are potentially as important as each other and character space is at a premium, which ones should be demoted to the Main Keywords Box?

    At the end of the day one might very reasonably conclude that this facility is best ignored and use [Lord Admiral Horatio Nelson] instead, as this requires less valuable space in the Essential Keyword box. So in practice, Exact Match may not be a very usable function simply because of the space that it will require.

    Weak Match; sound the most useful function and could be a major breakthrough, but that is assuming that it works just as well for any combination of words out of a much larger selection, as it does for just two words (the examples given). E.g./ [big large old ancient medieval stone building] will permit:- Large Ancient Building to be pulled up just as effectively as Big Stone Building, etc., and with the same ease and status as "Exact match" combinations of these.

    If this is the case, then all keywords relevant to a part, or subject, or concept in a picture could be square bracketed, so that keywords for each subject, concept, etc., are kept separate from those for others. By this I mean for example, 10 words in one set of brackets, 7 in another, 18 in another and so on, and of course square brackets within square brackets e.g./ [big large old [[Wesser Renaissance] style] stone building].

    However, this mode must result in no degrading of the effectiveness of each keyword, or of each combination, for it to be useful. Unfortunately the description of the working of this function could do with a bit more elaboration - at present we do not know if the search engine, will only cope with two words, would be seeking to link all words in the bracketed 10, 7, 18 words and be reluctant to take smaller combinations from these, and so whether it would effectively result in words in strings or double brackets being downgraded. Because if this is the case, then one is back into having to keyword, whether it be with [ ] or " " as per the silly example of Nelson above, and if one is forced into permutations and combinations, the whole business becomes absurd.

    Do Not Stem; sounds fine, but in practice could end up meaning that this symbol has to be placed in front of just about every other word! It is amazing what bizarre word stemming AlamyMeasures (Beta) highlights! And trying to pre-empt word stemming is almost impossible.

    Ordering; ref comments above for proximity.

    Referring now to the actual picture examples; both are what might be described as almost "single subject pictures" and so are relatively easy to keyword and as such are bad examples.

    Example 1 was the more "multi-subject" of the two, but where were the keywords relating to:- "Caucasian or white", "blond hair" (are these models? Odd that the father is dark and neither of the kids are?), "Lifestyle", "fitness", "healthy living", "diet", perhaps and where is the geographic location detail? The picture is patently taken on a beach and not in downtown Cape Town, so where is it? One cannot be vague like this when captioning most pictures.

    I also wonder why "mature men" and "mature women" (both plural) are used as main keywords when there are obviously only one of each i.e./ one man and one woman! And why families (plural) when it looks as if there is only one family? However, we all do things differently, and so I mustn't be too picky, though suggest that CTR would take a knock because of these.

    As already said, the image is a rather "single subject" one and so does not show how to deal with multiple subjects and concepts within a picture. How does one deal with say "architectural considerations", "historical considerations", "weather/time of day/season/mood considerations", "emotive/feeling considerations", all of which might sell a particular image, when one has limited keyword space? Which does one put to the front and which does one relegate to the back, and why? This is not dealt with in these examples.

    Also a "single subject" image is much less likely to have keyword combinations effecting AlamyRank CTR. However in the first Alamy example, the man does not appear to be wearing a "dress shirt", nor the woman a "beach dress", neither are "brother and sister hugging" or "siblings embracing", or "father hugging daughter", and nor are they a "coloured family", or have "coloured parents/father/mother/children, etc"! And I am sure that word stemming will very unhelpfully throw up many more of like type even with this "single subject" image, though it is with "multi-subject/concept" images that unintended linkages between search words really take off with a vengeance!

    This is Alamy's example, and I hope that from this it Alamy now see that however hard one tries, the dear old "mindless search engine" and "word stemming" will find holes in one's best efforts, and one's CTR will suffer as a result. And it is not just that one image, it is the entire collection under that pseudonym that takes the knock. And that is before researchers fail to specify in their search that they actually want their family in the city of Cape Town and not on some unknown beach, and sitting instead of standing, or whatever is their fad!

    One cannot win at this, because the researchers' contribution is not known at the time of keywording, which is why I do not agree with the AlamyRank concept of assuming that images which are called up and subsequently not zoomed, are badly keyworded and by implication "irrelevant". This is a totally false assumption.

    Alamy is right to attempt to tighten up keywording and I feel that the [weak link] idea has the potential to be a real breakthrough - it all depends on how it actually works in practice - but do feel that the current mode of calculating CTR is deeply flawed and needs looking at again with a view to measuring that which is positive about any one search, rather than measuring that which is negative and is so through no fault of the contributor.

  31. 12 February 2008 at 05:10 Ellen mcknight

    July = Julie

    I couldn't figure out for the life of me why my photo of a scene painted by Andrew Wyeth (in July 2005) was coming up under searches for Julie Andrews!

  32. 13 February 2008 at 08:10 IanMurray

    dung/dungeness

  33. 16 February 2008 at 22:40 Jon lewis

    Why when searching do images of a particular subject with NO KEYWORDS come before my or other contributors images with comprehensive keywording?. We have bothered to do the work to keyword they haven't we are penilisedfor doing the keywording they get further up the search results.
    I am confused.

  34. 16 February 2008 at 22:41 Jon lewis

    Why when searching do images of a particular subject with NO KEYWORDS come before my or other contributors images with comprehensive keywording?. We have bothered to do the work to keyword they haven't we are penilisedfor doing the keywording they get further up the search results.
    I am confused.

  35. 21 February 2008 at 16:57 jenny tonkin

    Would it be worth putting another couple of tick boxes in the 'manage my images' section so that you could define which mode the photo was taken in i.e portrait or landscape?

  36. 22 February 2008 at 18:59 Shaun

    Pompei in Italy returns Pompey (sort for Portsmouth)

  37. 06 March 2008 at 02:05 Jacquues Jangoux

    Question to Leslie Garland (post 29)

    "Inspection of my own AlamyMeasures (Beta) results indicates that by far and away the vast majority of searches are made using only two, yes two search words!"

    Leslie, how did you find out? By comparing your keyword list to the image? Or does Alamy mention somewhere which keywords have been searched?

  38. 14 March 2008 at 10:11 Krystyna Szulecka

    metallic
    metal

  39. 14 March 2008 at 10:11 Krystyna Szulecka

    metallic
    metal

  40. 11 April 2008 at 14:01 shaun

    generator
    generations

    PS - there are loads of suggestions here - is anything being done with them?

Add your own comment
Means that we require this information. We respect your privacy and store your details securely. See our privacy policy for more.
Security code picture
No picture? Reload page.
No picture? Reload page.