2.7K Messages
•
47K Points
Keyword Searches: IMDb has changed its display format for search results
As of a few minutes ago, the IMDb keyword search results changed significantly when I view in my browser.
I will discuss the new display in this post, and will also ask three questions (in bold-face font below).
Below are a couple screenshots of what the search results look like now.
https://www.imdb.com/find?s=kw&q=argue+couple&ref_=nv_sr_sm
The search results are presented in a more "modern" format on a white background, with larger, bolded text for the keywords.
One big change is that the number of titles for each keyword is now presented on a separate row from the keyword itself. This will hurt my ability to post lists of duplicate keywords for mass mergers. Previously, I could easily copy and paste keywords to assemble lists like this:
arguing-couple (15 titles) --> couple's-argument (2 titles) --> argument-between-couple (75 titles) --> couple-argues (27 titles)
knife-held-to-someone's-throat (40 titles) --> knife-held-to-one's-throat (20 titles) --> holding-a-knife-at-someone's-throat (22 titles) --> holding-a-knife-to-someone's-throat (11 titles) --> hold-a-knife-to-someone's-throat (3 titles) --> knife-held-to-throat (444 titles)
Using the new display, such a list would look something like this (matching the keywords in my first example shown above):
- 15 titles
- 2 titles
- 77 titles
- 30 titles
I could manually edit such a list to present it in the old format, but that would take a lot of work. Which brings me to my first two questions.
1. Is it possible to opt out of the new display format for search results?
2. If not, could the new display show the number of titles in parentheses in the same row as each keyword, like it was before?
One other major change: for keyword searches on popular words, the old display would include up to 200 results. The new display only includes 25 results at a time, and you have to keep clicking "More popular matches" to show more. Here is an example link to show what I mean. Having to click that button 7 times to get the same 200 results as before is less than ideal. On the more positive side, if you click the button enough, you can display way more than 200 results. I tested it using this link and got up to 1,200 results, which is likely the limit. But that required clicking the "More" button 47 times. Which brings me to my third and final question.
3. Can the new display be modified to show up to 100 keyword results as the default (instead of 25 results)?
Displaying up to 100 results would really help with the usability/functionality of the new system. I would even settle for 50 results, but I would prefer 100 results. 25 results is way too small as a default, and will require a lot of button clicking just to get a good understanding of what is going on with popular words within keywords.
Col_Needham
Employee
•
7.3K Messages
•
179.9K Points
2 years ago
@keyword_expert Thanks for the feedback. To answer your questions:
1. No sorry, the new experience is gradually being rolled out to increasingly larger numbers of customers. Once it reaches 100% the old search software will be retired.
2. No sorry, this is the new display format. Since your keyword merge requests are manually reported and manually processed anyway, just do not worry about how they look.
We will report the result sizes issue to the appropriate team. There’s a smaller initial result set so as not to overwhelm people with results when it is likely their desired result is in the first set anyway, however, once a customer presses a “more” link there’s less reason to keep adding results in only 25 item chunks.
3. See above :-)
Hope this helps.
3
0
keyword_expert
2.7K Messages
•
47K Points
2 years ago
I figured out a way to get back to the old display format -- at least for now. Simply logging out of my IMDb account, and then logging back in, works.
Here are screenshots of the old display format, for comparison to the screenshots of the new display I posted above.
For the "knife throat" search, when comparing the old version to the new version, it's clear that not only is the format different, but the ordering of the search results is different, too. The old search was more likely to incorporate different variations of words (and similar words) into the results and display them earlier in the results, while the new search is more strict about searching for the exact words entered and displaying those first in the results.
I believe this is another reason why the initial results should display more than 25 results. With the new search format, if someone searches using a word like "decapitation," they won't see results like "decapitated-corpse" until much later than the first 25 results. I wish I could show screenshots of that for comparison purposes, but when I logged back in, I am now back to the old search results display (which I do prefer overall -- the only thing I like about the new display is the option of displaying up to 1,200 results).
2
keyword_expert
2.7K Messages
•
47K Points
2 years ago
The one good thing I will say about the new keyword search display is that it now picks up newly created keywords within just a few minutes after the keywords are created.
Previously, the search system would only pick up newly created keywords once per week, on Sundays.
Now, they are picked up almost immediately. They appear as "0 titles" on the initial search results display, until the number of titles is updated on Sundays.
But this is about the only good thing I can say about the new search. I still much prefer the old search, which displayed up to 200 results at a glance, gave the number of titles in the same row as the keyword, took up less space on the screen per keyword, was easier to read and use, and incorporated a "smarter" search engine.
1
keyword_expert
2.7K Messages
•
47K Points
2 years ago
Could somebody please take this thread off of "Solved" status? I don't regard most of the issues discussed herein as solved.
0
keyword_expert
2.7K Messages
•
47K Points
2 years ago
@Col_Needham @Jon
I discovered some more problems with the new keyword search engine, involving "exact matches" of the keywords searched for.
I will illustrate these problems using the exact keyword search "brother relationship" as an example.
Let's say I search for "brother relationship" because I am interested to see if that exact keyword exists, and also because I want to see if similar keywords exist.
Here is that search, and below is a screenshot of the current top 11 results: https://www.imdb.com/find?s=kw&q=brother+relationship
Note that the exact keyword I searched for, "brother relationship," does not appear until #10 in the results.
And strangely, "brother brother brother relationship" appears as #1 in the results, despite its relatively low popularity compared to the very popular "brother brother relationship" (which appears as #2).
Apparently the search engine assigns extra weight to the repeated prevalence of search words within the keywords themselves. This doesn't feel like the right way to power the search engine.
Here is how I would fix this. If the keyword searched for has an exact match in the current keyword system (referred to as "exact matches" (plural) in the new keyword search engine), I would always list that exact match as the #1 spot in the "popular matches" results.
For example, for my sample search "brother relationship" there is currently an exact match keyword for "brother-relationship" so I would list that exact keyword first in the "popular" results.
As for spots #2, #3, etc., I would power the keyword search engine more like the old engine than the new engine.
The old engine "smartly" assigned extra weight to the prevalence/popularity of similarly worded keywords, placing extremely popular keywords higher in the results, above extremely unpopular keywords that might begin with the same exact wording.
In contrast, the new engine returns too many keywords that begin with the exact words searched for, without regard to popularity, thus returning too many keywords in the top 25 results that are currently only assigned to 1 title.
And as previously discussed, the new engine for some reason gives extra weight to keywords that repeat the term searched for, in this case placing "brother-brother-brother-relationship" as #1 in the results. That makes no sense.
The old engine also did a better job of flipping the order of the terms searched for, in order to provide more relevant and/or popular results higher in the list.
For example, if you do a keyword search for "child murder" the new engine does not display what you are (likely) actually searching for: the very popular keyword "murder-of-a-child," until #26 in the results (which, of course, is not within the first 25 results, which also involves the other serious problem I have tried to bring to your attention -- that 25 results is too few).
https://www.imdb.com/find?s=kw&q=child+murder&ref_=nv_sr_sm
Again, "murder of a child" is #26 in the results. Thus, if you search for "child murder," you will literally not even see "murder of a child" in the so-called "popular" results returned by IMDb. This is a problem.
I have to wonder whether IMDb is powering its new search engine with the exact same engine when searching for "popular" keywords as when searching for "popular" titles and celebrities. It does seem that way to me. This would be a mistake, since there are important differences with keywords.
For example, keywords are unique in that it is impossible to have exact duplicate keywords. There can be multiple movies named "Brother Relationship" and multiple people named "John Smith" but there can only be one keyword named "brother-relationship" and therefore only one exact match when you search for that keyword. The unique things about keywords really need to be factored in to the search engine.
I wish there were a way to access the old keyword search engine and test it side by side with the new engine to provide even better examples of why the old engine was overall superior to the new engine. Hopefully the issues discussed in this thread will give you a sense of some of the many problems with the new engine, which does not appear to have been properly vetted.
0
keyword_expert
2.7K Messages
•
47K Points
2 years ago
Here is another example to illustrate one of the problems with the new search engine for keywords.
When I do a keyword search for "killed," popular keywords like police officer killed, returning character killed off, actual animal killed, characters killed one by one, man killed, innocent person killed, shot and killed, etc. do not start to appear until result #411.
That's because the new search engine first returns all keywords that literally begin with the word "killed," including hundreds of keywords currently only assigned to 1, 2, or 3 titles.
And yet this new search engine purports to be a search for "popular matches." Obviously that is not true.
I am pretty confident in saying that the old search engine would have done a better job of returning these actually popular keywords (bolded below) higher in the list of results.
Search "killed"
Keywords
(edited)
0
keyword_expert
2.7K Messages
•
47K Points
2 years ago
Here is another example search to illustrate one of the problems with the new keyword search. When I search for "brother sister relationship" that exact match is listed #2 in the list, after a much less popular keyword.
If there is an exact match for any keyword, that exact match should always be given the #1 spot in the results. But this concept is limited to exact matches (of which there will always be only 1 or 0 for any given keyword search, since every keyword is unique).
1
keyword_expert
2.7K Messages
•
47K Points
2 years ago
As further proof that there is something really "off" about the new keyword search engine, here is another example.
When you search for "one day time span" (with a space in "time span") the very popular and nearly identical result "one day timespan" is literally the last keyword listed in the search results. And because the new display takes up two rows per entry instead of one row, when viewing on most computer screens, that last search result won't even be visible unless you scroll down.
Search "one day time span"
Keywords
(edited)
0
keyword_expert
2.7K Messages
•
47K Points
2 years ago
Here is another weird situation illustrating the shortfalls of the new keyword search algorithm.
Why is the exact item searched for not listed until the fifth result?
Search "reference to laura palmer"
Keywords
0
jeorj_euler
10.7K Messages
•
225.4K Points
2 years ago
The following is a bookmarklet (or script code prefixed with a out-syntax scheme identifier) which, when used on an IMDb keyword search results page, will convert the format to the way effectively it used to be.
If you know what a bookmarklet is, then you already know what to do, otherwise to "install" the tool, you would highlight the aforementioned source code (a JavaScript URI), copy it (to the clipboard) and paste it into URL field of a browser bookmark, whether already existing or newly created.
To use a bookmarklet, you "visit" it while having the particular webpage (in all of the Web) as the active tab in your browser. You don't need to understand JavaScript or any scripting language to use a bookmarklet.
The properly-encoded URI for that same source code is (function(p,m,e,i,n,t){for(n=e.length,i=0;i<n;i++){e[i].style.display="inline";t=e[i][m]("a"),t[p]=t[p].replace(/%20/g,"-");if(t=e[i][m]("ul%20li%20label"))e[i][m]("ul").outerHTML="%20("+t[p]+")"}})("innerHTML","querySelector",document.querySelectorAll(".ipc-metadata-list-summary-item__tc")); by the way. I previously presented this with line breaks and indentation to make it easier to read and thereby easier to understand, which in turn helps a semi-experienced individual screen it for malicious script behavior and errors in the flow of control that could unintentionally cause problems with resource allocation on any computer on which it is executed, meaning that trusting me would not be requisite for understanding that the script is safe.
I don't have a script for generating a longer or comprehensive listing of keywords in a search results page, as of this time.
4
keyword_expert
2.7K Messages
•
47K Points
2 years ago
Another example: the exact keyword searched for, "reference-to-christ," does not appear in the results until result #48.
Search "reference to christ"
Keywords
0
jeorj_euler
10.7K Messages
•
225.4K Points
2 years ago
The following is a bookmarklet (or script code prefixed with a out-syntax scheme identifier) which, when used on an IMDb keyword search results page, will convert the format to the way effectively it used to be.
If you know what a bookmarklet is, then you already know what to do, otherwise to "install" the tool, you would highlight the aforementioned source code (a JavaScript URI), copy it (to the clipboard) and paste it into URL field of a browser bookmark, whether already existing or newly created.
To use a bookmarklet, you "visit" it while having the particular webpage (in all of the Web) as the active tab in your browser. You don't need to understand JavaScript or any scripting language to use a bookmarklet.
The properly-encoded URI for that same source code is javascript:(function(p,m,e,i,n,t){for(n=e.length,i=0;i<n;i++){e[i].style.display="inline";t=e[i][m]("a"),t[p]=t[p].replace(/%20/g,"-");if(t=e[i][m]("ul%20li%20label"))e[i][m]("ul").outerHTML="%20("+t[p]+")"}})("innerHTML","querySelector",document.querySelectorAll(".ipc-metadata-list-summary-item__tc")); by the way. I previously presented this with line breaks and indentation to make it easier to read and thereby easier to understand, which in turn helps a semi-experienced individual screen it for malicious script behavior and errors in the flow of control that could unintentionally cause problems with resource allocation on any computer on which it is executed, meaning that trusting me would not be requisite for understanding that the script is safe.
I don't have a script for generating a longer or comprehensive listing of keywords in a search results page, as of this time.
0
jeorj_euler
10.7K Messages
•
225.4K Points
2 years ago
I just edited and deleted Sprinklr comment 63a8ffb88a78b24108718f80, so Sprinklr reply 63ae62c548a8017448664f11 by keyword_expert has become invisible to everybody, even though a record of it has been kept and still shows up in my notifications. I'm noticing some glitches in the way that the forum software filters the contents of posts. For instance, whenever the words "top", "bottom", "left", "right", or "display" are followed by a colon and a quotation mark, the filtering system will remove the whole word, the colon and that quotation mark, along with any white space characters between the word and symbols. The aforementioned words are Cascading Style Sheet (CSS) description properties, and may appear in HTML "style" tags or in the HTML "style" attributes of respective HTML element tags.
4
keyword_expert
2.7K Messages
•
47K Points
2 years ago
Some new changes have been made very recently to the keyword search engine that now place popular similar keywords at the top of the list.
For example, in my post of 21 days ago, I reported that the search for the word "killed" did not return "police-officer-killed" and similar keywords until result #411. Today when I do the same search, "police-officer-killed" is literally at the top of the list, ranked at #1.
Search "killed"
Keywords
Another change I have noticed is that when you click the "exact matches" button on keyword searches, it will now sometimes return multiple results (not just one result), because it now searches for exact matches without regard to spaces and punctuation.
For example, a search for "videoconferencing" returns both video conferencing and videoconferencing as exact matches.
A search for "voicemail" returns both voice mail and voicemail as exact matches.
A search for "esp" returns both esp and e.s.p. as exact matches.
All of these are good improvements.
However, back to the first change (placing similar popular keywords at the very top of the list), I actually think this change has gone too far in that direction.
I would still recommend always placing exact matches at the very top of the search results list regardless of popularity (especially since we are only talking about one or two search results).
And I would still recommend placing keywords that are extremely close in wording (without regard to popularity) to be placed next, after the exact matches.
Then the popular similar results should be mixed in with the results that begin with the same words.
An appropriate balance should be struck. The new engine is closer to finding that balance, but it hasn't quite been found yet.
To illustrate my point, consider the following search. Currently, when I search for "trapped in a moving car," the keyword trapped-in-a-driverless-moving-car does not appear until result #35 (and because the results still only display 25 initial results as default, this keyword is not even visible at first). "trapped in a driverless moving car" is so close to "trapped in a moving car" that I would expect it to be higher in the search results. (And I also still think that initial results should include 50 keywords rather than 25 keywords).
Search "trapped in a moving car"
Keywords
(edited)
1