intentPar
intentPar is a term used in conversational AI and natural language understanding to refer to the probability distribution over a set of user intents for a given utterance. It arises from an intent classifier that assigns scores or probabilities to predefined intents such as book_flight, weather_query, or greeting. The resulting probability vector, sometimes called the intent distribution, is used by the dialogue manager to select the most likely intent, trigger corresponding actions, or decide on fallback behavior when confidence is low.
In typical deployments, the classifier outputs either probabilities or logits that are converted into probabilities by
Applications of intentPar include routing to the appropriate skill or handler, deciding whether to ask a clarifying
Variations exist, such as storing log-probabilities instead of probabilities or applying temperature scaling to adjust confidence.