AI Search - Knowledge article relevancy

[Tony Croft]

Hi All,

I am currently configuring Virtual agent and working on AI search.

Is there a way to improve the AI search accuracy/relevancy of the knowledge articles that are offered after a query has been entered. 

If I enter garbage into the query i.e "adffggbhd" then the VA correctly routes to the fallback topic which is great.

If I enter a query that is totally irrelevant and doesn't have any knowledge articles i.e "My apple is a banana" then it comes back with knowledge articles that have no relation or relevancy to the query when it shouldn't return anything. In this example the top knowledge article returned was instructions for "Mapping a shared drive". I'm certain that article contains no references to apples or bananas so dont understand the logic it used.

Ideally if no relevant knowledge articles are found then it wouldnt display them and go to the fallback. Is there a way to configure AI search to only show relevant articles. 

With NLU there is confidence thresholds which can be tweaked to only show valid intents but with AI search I dont see a way of doing something similar. 

I have taken a look at the AI search tuning article but not sure if this would help in my scenario

Thanks

[Aaron6]

Hi all - same exact question.

@Tony Croft FYI contextual search results have a 'score' property that you can use, but I cannot find anything for AI Search. Did you ever solve for this?

[Bernard Esterhu]

I'm also looking for an answer to this question. We are integrating AI search in the Virtual Assistant, and I would like to prevent low accuracy results from being displayed to the user.