Because NLP uses a more complex algorithm than keyword matching, it can be difficult to adjust your training phrases manually without causing unexpected results.
You can:
- Make sure your passages define a maximum of one question at the global and conversation context levels.
- Use a review to monitor the accuracy of your NLP questions through analytics and generate recommended adjustments to your training phrases.
- Use regression tests to ensure changes to your training phrases don't negatively impact your chatbot.
- Scan your chatbot for issues with your questions.
You can also create a TrueIntent usage with a rewrite utterance task to rephrase questions from your chatbot users that are difficult to match.