Open jeffpaul opened 4 years ago
Bah, I quickly created this after creating #147 and completely forgot that it's Watson NLU for the language processing. Still, the idea holds of allowing us to swap Watson NLU out for a custom modeling service that could learn from any edits that a site owner makes to tags. I'll update this issue to be more accurate to ClassifAI's currently service providers in a bit, mea culpa.
Watson NLU announced custom classifcations now in beta and custom category models as well (which are both part of their broader NLU custom model offerings), so this is something that could begin to be explored should the market/community see value in custom text classifications.
Is your enhancement related to a problem? Please describe. We currently utilize Azure's Text Analytics for the Language Processing functionality, but you're forced to accept what the general Text Analytics model has learned to get feedback in the form of tags. A parallel option could be to utilize the Language Understanding service and further train that model to be site-specific. This would seem to provide a more precise set of language processing results unique to a specific site, though a downside could be pricing of Language Understanding compared to Text Analytics.
Describe the solution you'd like
Designs n/a
Describe alternatives you've considered Keep ClassifAI as-is with Text Analytics and instead work to add additional functionality instead of more complex functionality as described here.
Additional context n/a