Fuzzy Search

[embyuser24 51]

Hi,

 

The current implementation of Emby search is very basic. Based on what I can tell, it only matches exact strings in the same exact order.

 

Example: "Mission: Impossible"

In this example the following queries will return a positive match

• Possible
• Impossible
• Mission
• Mission: Impossible

And the following queries will not return anything at all (The worst offenders are highlighted)

• Mission Impossible
• Missoin: Imposible
• Imposible
• Impossible Mission
• Mission : Impossible
• Mission     Impossible

Another example would be "Mr. Robot" where if the dot is missing, nothing is returned or in general terms any variation of any string that is not an exact match will not return anything at all. Let's not even attempt to try a title like "Agents of S.H.I.E.L.D.".

 

This makes the search difficult to use and at times even frustrating. As the examples above show, the major theme is special characters and punctuation and they can probably be bypassed by just ignoring them, which might or might not cause other problems in the process. This problem also affects titles that have letters with diacritical marks or glyphs such as "Amélie" and "Æon Flux", if you can't type them exactly, you can't find them.

 

In my opinion a better attempt would be to use a difference string matching algorithm such as Levenshtein distance, you can then define an acceptable threshold and return the closest matches.

 

I found a couple of C# projects that implement this and might be useful as a starting point. 

 

• FuzzyWuzzy, a port of Python project in C#. https://github.com/BoomTownRoi/BoomTown.FuzzySharp
• FuzzyString. Has some additional string comparison algorithms too. https://github.com/kdjones/fuzzystring

 

Thanks.

 

[Luke 40063]

Hi yes, it's something we're interested in for the future. thanks.

[legallink 187]

+1 on this.

[legallink 187]

+1 on this.

[invertedknife 13]

+1 on this

[adrianwi 253]

+100

[embyuser24 51]

The new search improvements in v4.0 are great and very welcomed!

 

Are there any plans for implementing nearest match searches in cases of misspelled words/phrases?

[Luke 40063]

It's possible for the future. Thanks !

[Bottles51 70]

Can we please have the apostrophe excluded in the web browser searches. It is easy to forget adding one in especially when pasting in a title. You may have the title and forget to add the apostrophe and not realize it.

[Luke 40063]

HI, yes this is something that needs to be improved. Thanks.

[Jägs 85]

I'd just like to expand the scope on this a bit since the original request was video specific, but with improvements made to the music side of the shop, it'd be great to include artist and title (or album) in the search as well, as I outline here:

 

[Luke 40063]

Taking another look at this, this feature request was created prior to Emby Server switching to full text searching in it's database, and as a result, many of the examples in the first post have been solved for the past several years. Even the diacritical marks have long been solved where Amélie can be located by searching either Amélie or Amelie.

The open issues are currently misspellings (fuzzy), and certain characters such as single quotes and periods that require the person searching to enter their search term exactly as it appears in the database.

In the upcoming 4.8 server release, we'll handle these characters better so that it won't matter how you're searching.

[Jägs 85]

Just so I'm clear:  the improvements to search do not include the ability to search artist/composer + title (e.g., "Miles Davis Autumn Leaves"), correct?

[Luke 40063]

3 hours ago, Jägs said:

Just so I'm clear:  the improvements to search do not include the ability to search artist/composer + title (e.g., "Miles Davis Autumn Leaves"), correct?

Correct, however there are some other nice improvements you may enjoy.

[donkorleone 0]

I'm running 4.8.0.46, thanks for the improvements!

Don't want to keep this open forever, but I'm adding my use case. For a video with the name, say "Vacation.Video.2023-07-13.Rome.1" if I search for "Rome," then I don't get back the item. I can technically go and bulk edit metadata, but I have hundreds of these files and lots are with a similar naming convention.

[Luke 40063]

4 minutes ago, donkorleone said:

I'm running 4.8.0.46, thanks for the improvements!

Don't want to keep this open forever, but I'm adding my use case. For a video with the name, say "Vacation.Video.2023-07-13.Rome.1" if I search for "Rome," then I don't get back the item. I can technically go and bulk edit metadata, but I have hundreds of these files and lots are with a similar naming convention.

What is the title of the item in the server?

[ryancey 33]

Any news on this feature? Looks like I hasn't made it to the 4.8 release and yet seems pretty expectable nowadays on a library manager. Thanks!

[Luke 40063]

9 hours ago, ryancey said:

Any news on this feature? Looks like I hasn't made it to the 4.8 release and yet seems pretty expectable nowadays on a library manager. Thanks!

Hi, yes we are continuing to work on improving search. Thanks for the feedback.

[ylatsj 2]

Example: "Mission: Impossible"

In this example the following queries will return a positive match

• Possible
• Impossible
• Mission
• Mission: Impossible

 

The fact is not like what it is described!!!!

impossible, possible won't return any result.

only miss,mission will.

 

 

[CummingCowGirl 28]

On 2/20/2024 at 12:33 PM, Luke said:

Hi, yes we are continuing to work on improving search. Thanks for the feedback.

Search has needed to be fixed for a very very very long time. When doing a search within a specific library Emby should be only looking for results in that library. When doing a search in settings Emby should only be searching for results in settings. 

[mammt 4]

+1, since 2018 and it hasn’t been done yet its crazy.

Some Examples:

• Searching for "spiderman" yields no results, but searching "spider-man" works as expected.
• Searching for "starwars" shows no results, but "star wars" displays all relevant content.
• Similarly, if i search for "Storia d inverno" or "Storia inverno" it shows no result, but "Storia d'inverno" it shows the movie correctly.

Edited December 29, 2024 by mammt

[TZTZoro 25]

for the people who want to have a faster implementation then when the devs do it, maybe check out this project https://gitlab.com/DomiStyle/jellysearch

disclaimer:
i did not try it out but i know that jellyfin and emby api is very similar structure with minimal differences, so someone who is willing might easily port it to emby

[TZTZoro 25]

and to change the feature request, could instead of it just being a fuzzy search being a phonetic based fuzzy search, it might improve performance to my knowledge and improve search results in my opinion

smote smite smith, are all 1 edit distance away but when i type smite i most definetely do not want smith, okay you can still consider it a typo but on a qwerty or zwerty based keyboard your finger was quite off from the original letter, and i believe that a phonetic based fuzzy search would improve for things like common spelling errors, or if creating a hybrid system, could there be a ranking system in the backend like this is most likely what you want but if you typed it very badly you might also have wanted this

[mammt 4]

any news on this?

[Jägs 85]

While we're at it, it would be great on the music side if we could search by artist and title, as I note here: