When you use NLU (Natural Language Understanding) to help predict the topic as users interact with the Virtual Agent at some point you might find yourself needing to improve, or tune, the NLU model to better understand the context or lingo used in your organisation. One area that can seem quite daunting is to use the vocabulary to help train the system to learn the language of your data. Examples are acronyms typically used within your organisation but not in common language, or abbreviations used by users in the organisation.
It is important to use NLU vocabulary in a model as it helps to improve the model's intent prediction accuracy. For example, if a word in a users utterance (typically an inquiry or request) is an acronym or is specific to one domain, the system may be able to predict its intent from its context within the utterance. However, when you define the vocabulary details for the word using a vocabulary item, you can help raise the model's confidence level and strengthen its inference capability.
In this knowledge base article you can find a document with a set of good practices that will help you understand when and how to make updates to your NLU vocabulary. Although it is updated for the Quebec release, it also applies to the Paris release for the most part.
