A Natural Language Understanding model operates on similarity. With every training, the model analyzes similarities and differences among phrases in a training dataset and classifies them into groups that share common characteristics, mostly sentence structures and word frequencies.
That is how the model learns to categorize user inputs. Every time a new sentence is provided, the model compares it to the categories it has found in the training dataset and counts the probability of assignment to each class (intent).
Therefore it is crucial to create categories as different and separate as possible.
Updated over 1 year ago