Learn how to train your bot to understand what the users are asking about and how to extract structured data from user messages.
Since human language is not that straight-forward and since there are multiple ways to say the same things, the bot should be provided with enough examples for each conversation and entity.
The process of providing these examples and updating them to improve the understanding is called Training.
Training your bot to understand user input is essential.
For a good explanation of the way NativeChat Natural Language Processing works, check the What is Natural Language Processing (NLP)? post in our blog. It gives a lot of useful examples and will be a very good start for learning how to train your bot successfully.
After you’ve covered the basics above, you can get into more details for using the different NativeChat NLP features and capabilities:
- Conversation triggers that will help your chatbot identify the correct conversation from the Cognitive Flow that needs to be started.
- Entity training that will teach your bot to recognize Entities (like persons, companies, locations etc.) from a conversation with a user. The training can be either static defined manually or dynamic defined by a web service call to train your model from existing data.
- Question answering - used when the bot can directly provide an answer to the user question or phrase. For example, when the user says bye, thanks, or What are your working hours?.
- Problems check - find conflicts and mismatches in the training examples provided.
- Understanding assessment - annotate user phrases to enrich FAQ capabilities.
- Test set - see and modify phrases added through Understanding assessment.
Built-in Training Data
Each NativeChat bot is created with some pre-trained conversations (welcome, help, restart) and Small Talk (reactions to thanks and bye utterances). This covers the most basic scenarios and is needed for the bot to work correctly out-of-the-box.