Hai,
How to handle the misspelling situation.
I have a FAQ question like what is global search?
and if user gives global seaarch? means it throw fallback . can you please suggest arrest this scenario?
Hello Vithya,
Have you tried enabling the auto-correct feature option under Natural Language → Thresholds & Configuration → Knowledge Graph for your bot.
Hai Deepak,
Thanks … Now its is rectified. but if there is a spelling mistake in small talk it how to handle those.
Hello Vithya,
For such cases you may want to create concepts and possibly add those misspellings there. I know this is not fool proof but possibly the only way. The other way could be if you have enabled the Few Shot model which uses the Kore’s native LLM model and might be helpful.
As a general rule, misspellings should NOT be added to training.
Any new words added to the training data are considered as being additions to the bot’s dictionary and hence are blessed as real words. The consequence of that is when the word appears in a user’s utterance it will not be corrected (because it is a “real” word), and that it becomes possible for other misspelled words to be corrected to that first misspelling (because it is now designated a “real” word).
The platform does perform spell correction, so if there are specific examples of misspellings not being corrected then please quote those examples.
Great callout Andy. I ll keep that in mind the next time we have somethig similar.