Methods and systems for training and deploying natural language understanding models

An electronic device, comprising:

• a memory comprising a natural language understanding (NLU) model;
• and
• a processor configured to:
• provide an utterance to the NLU model, wherein the NLU model is trained by:
• obtaining a dataset comprising utterances, wherein the utterances comprise one or more words;
• generating a list of intents by analyzing overlap and steepness values of word combinations in the dataset to identify core actions or topics to act as intents;
• grouping the list of intents into one or more domains;
• nesting intents sharing a common topic into a domain representing the common topic;
• stacking intents similar in meaning within the domain into a unified intent;
• generating a list of training utterances for each intent in the list of intents by threading words representing the intent through each training utterance corresponding to that intent;
• and
• modifying one or more parameters of the NLU model based on at least one of the list of intents, the list of training utterances, or the one or more domains;
• select, by the NLU model, an intent by first selecting a domain and then selecting an intent from the list of intents for the selected domain based on the utterance;
• and
• provide a response to the utterance based on the selected intent.

A non-transitory computer-readable medium storing instructions that, when executed by a processor, causes the processor to perform operations comprising:

• providing an utterance to a natural language understanding (NLU) model, wherein the NLU model is trained by:
• obtaining a dataset comprising utterances;
• generating a list of intents by analyzing overlap and steepness values of word combinations in the dataset to identify core actions or topics to act as intents;
• grouping the list of intents into one or more domains;
• nesting intents sharing a common topic into a domain representing the common topic;
• stacking intents similar in meaning within the domain into a unified intent;
• generating a list of training utterances for each intent in the list of intents by threading words representing the intent through each training utterance corresponding to that intent;
• and
• modifying one or more parameters of the NLU model based on at least one of the list of intents, the list of training utterances, or the one or more domains;
• selecting, by the NLU model, an intent by first selecting a domain and then selecting an intent from the list of intents based on the utterance;
• and
• providing a response to the utterance based on the selected intent.