How to improve smart tag accuracy

[CynthiaCo]

There are constant multiple errors in the smart tags generated for our photo assets. The AI also misses many obvious tags e.g. a close up of a flower has 8 smart tags, and none of them are "flower".  Or bizarre, nonsensical tags are assigned such as 'cowbell' 'lacustrine plain' or 'bobsled'. Is there a way to enhance its success rate? We've added 140 blocked smart tags to our system, however it remains time consuming to remove all the errors.

[Shashi_Mulugu · Accepted solution]

@CynthiaCo have you tried to train the assets by providing similar sample assets and tagged it properly? Are you in AMS or cloud?

I recently had one client with similar problem statement and we escalated this to adobe support via TAM as per client contract and adobe tech support trying to help on it.. please evaluate all options.