Natural language understanding clarifications
Topics: AI (Deep Learning), LLMO, Search Intent, Search Query Processing
This Google patent describes techniques for improving natural language understanding (NLU) systems by better handling cases where the system is uncertain about user intent. Specifically, it introduces a method for clarifying whether a user query matches a candidate intent when the confidence score falls just short of the main threshold but exceeds a secondary lower threshold. The system can then use various methods like user confirmation, historical mappings, or embedding space analysis to determine if the query should match the intent, helping to improve accuracy without constant model retraining.