Today we discuss the most fundamental task of all chatbots–intent detection. Intent detection is classified in the Training category of chatbot product features. It refers to the chatbot’s functionality of recognizing the correct intent from the customer’s inputs and then delivering the appropriate responses or executing the relevant tasks.
The enterprise user configures the intent detection process by defining one or more intents and then creating multiple utterances, or example sentences, for each intent. The chatbot platform will then build a machine learning model based on these user-defined intents and utterances. After the chatbot has been deployed, its intent detection functionality will compare the customer inputs against all the utterances and identify the intent whose utterance is the most similar to the customer input as the correct intent. The similarity score threshold which determines whether an utterance is considered close enough to the customer input can also be adjusted by the user.
For example, if the user has created an intent called “flight booking,” they will define multiple utterances for that intent to simulate what an end customer might say to the chatbot if they would like to book flights. These utterances might be “I’d like to book a flight from SFO to London,” “Can you get me an economy seat from Dallas to Portland,” or “Book me a flight on July 23 from LAX to Sydney.” After the chatbot has been deployed, if a customer input is similar enough to one of the utterances for the flight booking intent (“Can you please book me a flight from Seattle to Tokyo?”) then that intent will be triggered and the customer is taken to the next step in the flight booking flow. If the customer input is not similar enough to these utterances (“How much is the baggage fee for economy class?”) then the flight booking intent will not be triggered.
The screenshot above shows an intent and its utterances in an example chatbot vendor – Yellow.ai.