NLU Intent Clarification via Dual Threshold Evaluation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Natural language understanding (NLU) systems face challenges in accurately identifying candidate intents that are 'close' to satisfying a threshold likelihood value, leading to less accurate natural language processing.
Innovation Solution
The system processes user queries using an NLU model to generate candidate intents and intent scores, and determines whether the intent score satisfies a threshold likelihood value. If not, it checks an additional threshold value to assess if the intent score is 'close' enough, and performs further analysis or user querying to determine intent correspondence.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the system uses a strict threshold likelihood value to determine intent correspondence, then the decision-making process is simple and fast, but the accuracy of identifying candidate intents is reduced
Solution Approach 1:
The patent divides the threshold evaluation into multiple segments: a primary threshold for clear matches and a secondary threshold for borderline cases. This segmentation allows the system to handle different confidence levels differently, improving overall accuracy while maintaining manageable complexity through structured decision-making.
Solution Approach 2:
The patent applies partial action by performing additional analysis only on cases that fall within the secondary threshold range, rather than applying exhaustive analysis to all cases. This selective approach improves accuracy for borderline cases while avoiding unnecessary complexity for clear-cut situations.
2Measurement precision
If the system performs additional analysis for intent scores close to the threshold, then the accuracy of NLU improves, but the computing resources required increase
Solution Approach 1:
The system performs additional analysis only partially - specifically for intent scores that fall within the secondary threshold range but not for scores that clearly meet or fail the primary threshold. This selective application of additional analysis improves NLU accuracy for borderline cases while conserving computing resources for clear-cut cases.
Solution Approach 2:
The patent changes the parameter of threshold sensitivity by introducing a secondary threshold that is closer to the primary threshold. This parameter adjustment allows the system to identify and process only those cases that require additional analysis, optimizing the balance between accuracy and resource consumption.
3Measurement precision
If the system retraines the NLU model frequently to improve accuracy, then the NLU performance improves, but the time and resources required for model maintenance increase
Solution Approach 1:
The patent implements preliminary action by establishing a multi-threshold evaluation framework in advance that proactively identifies borderline cases requiring additional analysis. This preliminary structure improves NLU accuracy without requiring frequent model retraining, as the system is pre-configured to handle uncertain cases through additional analysis pathways.
Solution Approach 2:
The system uses feedback from the threshold evaluation process to determine when additional analysis is needed. By monitoring intent scores against both primary and secondary thresholds, the system receives feedback that triggers selective additional analysis only when necessary, improving accuracy while minimizing the need for frequent model retraining.
Data Source
AI summary
Techniques are disclosed that enable clarifying whether a user query corresponds to a candidate intent when an intent score (indicating the probability the user query corresponds to the candidate intent) fails to satisfy a threshold likelihood value but is “close” to satisfying the threshold likelihood value. For example, the intent score can fail to satisfy the threshold likelihood value but can satisfy an additional threshold likelihood value. Various implementations include generating the candidate intent and corresponding intent score by processing a natural language user query using a natural language understanding (NLU) model.