search

NCATSTranslator / Feedback

A repo for tracking gaps in Translator data and finding ways to fill them.
7 stars 0 forks source link

Change the behavior of this "Reasoning Agent" filter in the UI? #794

Open mbrush opened 1 month ago

mbrush commented 1 month ago

The Problem:

The current behavior of the "Reasoning Agent" filter is to show only Results from the selected ARA, but show paths/EPC info from all ARAs that also returned this result.

For example, when I look at the "Botulinum Toxin Type B" result for the query "What may treat Cerebral Palsy", and filter to show only ARAX results - I see many support paths that were not used by the ARAX agent (the circled paths below are those used by BTE, which also returns this result):

image

Also, other EPC that shows up in the results view or popup tables comes from BTE, not ARAX- even when an "ARAX results only" filter is on.

Proposed Improvement:

IMO, a much more useful behavior would be to also filter support paths (and other EPC info) to include only that used / provided by the selected ARA.

This is because the value of a "Reasoning Agent" filter to me is to be able to see not only what answers came form a given ARA, but more importantly what evidence and provenance each ARA used in doing so - which the current behavior does not support. The proposed change would also make it much easier for testers to understand where problematic paths / provenance is coming from, without as much back and forth between the UI and ARAX interface. This behavior would also helps educate developers, testers, and end users about how each ARA operates, and the types of paths/evidence they like to use.

Unless there is value to retaining Paths/EPC from all ARAs upon filtering for results from a single ARA, can we change the behavior of this "Reasoning Agent" filter?

dnsmith124 commented 1 week ago

I think the requested adjustment to this feature is very sensible, but in terms of prioritization I wonder how much value the Reasoning Agent filter has for a normal user compared to a tester within the consortium?

It would be a fairly complex expansion of the filter, particularly considering when multiple ARAs are enabled/disabled, and we'd need to decide how to display the excluded paths (if at all).

dnsmith124 commented 1 week ago

@Genomewide see above

gprice1129 commented 3 days ago

UI team needs to discuss internally before assigning to a release.

Genomewide commented 2 days ago

@dnsmith124 Couldn't this be covered in the same way we have talked about the other facets? Bring those that fit the criteria to the top and make the others opaque.

dnsmith124 commented 14 hours ago

@Genomewide that's correct, I had commented the above before we discussed as a team potentially overhauling the facet behavior with regard to paths