How does the intent detector work?
In order for a virtual assistant to understand what a user is asking, it uses an artificial intelligence solution called an intent detector. Every time a user types, the intent detector analyzes the text and identifies the user's intent.
But the virtual assistant only understands what it's been taught. In order for an assistant to understand the meaning of what a user is saying and provide the correct answer, it must be taught how a user can express a particular intent in the first place. There are hundreds of ways to ask the same question. Teaching an assistant is done by giving many examples of intents - user expressions.
Once the intents and examples are added to the virtual assistant's knowledge base, the assistant learns how to recognize them.
The user input is processed, converted into the form required by the algorithm, and passed to an artificial intelligence solution that calculates the score of all the intents known to the virtual assistant. When the scores of all intents are added together, the confidence level should generally be 1.
The confidence score for a particular intent indicates how confident the virtual assistant is that the text entered by the user matches that intent. If the confidence is high enough (by default the confidence level for detecting the intent is set to greater than 0.5, but you can change this threshold in Settings), the intent with the highest confidence rate for the user input will be detected. If all returned confidence rates are too low, the intent will not be detected at all.
