Order of matches found in knowledge base search is not correct

[ahoo1260]

Describe the bug
Our knowledge base has different string properties ( for example prefLabel, altLabel, label ). We like to get the entity with closest match in all of these properties, for example, if I search for "hand", the entity with altlabel "hand" should be shown on top of list.

To Reproduce
I have attached the project and the knowledge base.
import knowledge base local with IRI Schema
After importing knowledge base is done, search "hand"

Expected behavior
it should show "Hand structure (body structure)" on top of the results for search.

same as search results as Snomed browser https://browser.ihtsdotools.org/?perspective=full&conceptId1=85562004&edition=MAIN/2023-01-31&release=&languages=en&latestRedirect=false

Screenshots

But we expect this order:

Please complete the following information:

• Version and build ID: 26.8
• OS: Linux (ubuntu 20.04)
• Browser: chrome

Additional context
The project and kb file is attached

hand_project16118132736551889752.zip

vomiting_hand_graph.ttl.zip

[reckart]

@ahoo Thanks for the report - very insightful! I can reproduce the issue and am looking into a fix.

[reckart]

Is it important for you that Hand is at the top of the list or would it also be ok if it would be the preferred label of the concept, i.e. Hand structure (body structure).

The info in gray would not be shown normally - this is debug info in the development mode.

[ahoo1260]

Thanks. Preferred label is also ok! However, our annotators can be more confident if they see the exact match on the top of list ( which is hand ).

[reckart]

Would you then expect to see only Hand and not have Hand structure (body structure) in the list at all?

The label rendered on top of the text in the annotation bubble will always be the preferred label. I believe it would confuse the annotators if the item they chose is not the item that will be rendered in the text then.

I guess the best way would be to display in the item both, the preferred label and the label that had the best match....?Unfortunately also the more complicated solution to implement ;)

[ahoo1260]

Is it possible to see both? it is now the preferred name and the link to Snomed. If there is no room for everything to be shown, I believe the link can be replaced by the label that had the best match...

[reckart]

IMHO the IRI should also remain visible because some people enter the IRI directly and they should be able to compare what they entered to the items in the dropdown.

[ahoo1260]

very valid point. It is important that annotators can put the IRI

[reckart]

How about this:

[ahoo1260]

yes, looks exactly like what we want! Is there a reason than "hand bath" is the second match? and not "handicap"? I am trying to understand the sorting algorithm. Thank you

[reckart]

Ok, there is a longer story here :)

First, I had to change the settings in the project that you provided a bit - it was missing the settings to take the alternative labels into account:

In the screenshot above, I only had added http://www.w3.org/2004/02/skos/core#altLabel yet. Then I checked the Handicap entry in the KB and it turns out that it does not have a http://www.w3.org/2004/02/skos/core#altLabel, only a http://www.w3.org/2004/02/skos/core#prefLabel and a http://www.w3.org/2000/01/rdf-schema#label - so I added the ...prefLabel as well. With both the ...altLabel and the `...prefLabel as additional matching properties, we get:

In the gray debug info, you can see that Hand gun, Hand saw and Handicap all have the same edit distance to the query (levQuery=5 levQueryNC=4) - at which point the ordering falls back to alphabetic ordering and then comes before i.

[ahoo1260]

Thank you for explanation. It makes sense. I am very happy with the output

[reckart]

You can try this out in INCEpTION 27.0-beta-4: https://inception-project.github.io/downloads-beta/

[ahoo1260]

thank you Richard, I tested it and it works as expected. Will you make a release in the coming days?

[reckart]

I hope 27.0 can be released within the next 14 days - but I never give guarantees :) If for some reason it should get delayed, a 26.9 release can be added in between and that would also include the fix.

[reckart]

There will not be a version 26.9.

A release candidate 27.0-rc-1 is now available and you may consider testing it: https://github.com/inception-project/inception/releases/tag/inception-27.0-rc-1

[ahoo1260]

@reckart Is it accessible via docker ?

[reckart]

https://hub.docker.com/layers/inceptionproject/inception-snapshots/27.0-rc-1/images/sha256-81e705712099c32ea6d501730c5ed8dee5ca2fa096f4dd4ef82ea0f25ead5615?context=explore

[reckart]

For future reference - when configuring INCEpTION for Snomed, the following settings should be made:

• IRI Schema: OWL
• Root concepts: http://snomed.info/id/138875005
• Additional matching properties: http://www.w3.org/2004/02/skos/core#altLabel, http://www.w3.org/2004/02/skos/core#prefLabel