Kore.ai’s Bots Platform can allow for fully unsupervised machine learning to constantly expand the language model and the intent and entity recognition of the bot. However, we do not recommend a fully unsupervised ML approach. Pure machine learning based approaches for bot training can be very optimistic and can create a lot of false positives. This ultimately degrades the performance of the bot if full unsupervised ML training is enabled.
The Kore.ai Platform suggests a different form of “unsupervised learning” for NL in which all successful utterances (ie. the user utterances that were successfully recognized by the bot), are automatically used to expand the language model and re-train the bot. This includes user-provided confirmations of intents for conflicts. This approach enables automatic, constant training of bots as the ML model improves its accuracy and excludes failed utterances. Kore.ai believes this is a better approach to leveraging ML-based automated learning to improve the bot’s performance.