Intent Classification Training Using Distinct Intent Descriptions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing intent classification models struggle to accurately map user utterances to predefined intents due to challenges in distinguishing between independent and dependent intent descriptions, leading to suboptimal performance.
Innovation Solution
A method and system for training an intent classification model using intent descriptions, involving the generation of datasets with independent and dependent intent descriptions, and employing a ranker to calculate similarity and adjust the number of intent options, with retraining on portions of the dataset to enhance accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If intent descriptions are used to improve intent classification accuracy, then classification performance is improved, but the complexity of the training system increases
Solution Approach 1:
The patent segments the training process into distinct modules: a language model generation unit that creates intent descriptions, a dataset construction unit that organizes training data, and an intent classification model training unit that performs the actual training. This segmentation allows each component to be optimized independently while working together to improve classification accuracy without overwhelming system complexity
Solution Approach 2:
The patent introduces a language model as an intermediary component between the training data and the intent classification model. This language model generates intent descriptions that serve as a bridge, transforming raw user utterances into structured intent representations, thereby improving classification accuracy while managing system complexity through modular design
2Ease of manufacture
If independent intent descriptions are used, then training data generation is simpler, but intent classification accuracy decreases due to indistinguishable differences
Solution Approach 1:
The patent applies local quality by differentiating between independent intent descriptions (used for certain training scenarios) and dependent intent descriptions (used for other scenarios). The system selectively generates descriptions with or without contextual dependencies based on the specific training needs, allowing each type to serve its optimal purpose: independent descriptions simplify generation while dependent descriptions enhance distinguishability when needed
3Measurement precision
If dependent intent descriptions are used, then intent classification accuracy is improved, but training data generation becomes more complex
Solution Approach 1:
The language model acts as an intermediary that automatically generates complex dependent intent descriptions without requiring manual intervention. The language model takes user utterances and intent labels as input and produces context-aware intent descriptions that capture relationships between different intents, thereby achieving high classification accuracy while keeping the data generation process automated and manageable
4Adaptability or versatility
If the number of intent options is increased, then classification coverage is improved, but model performance deteriorates due to difficulty in distinguishing between options
Solution Approach 1:
The patent replaces traditional manual intent description writing with an automated language model system. This substitution enables the system to handle a large number of intent options by automatically generating distinctive descriptions for each intent, maintaining model performance even as classification coverage expands to include more intents that would be difficult to manually differentiate
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The system for training an intent classification model using intent descriptions of the present disclosure may generates a dataset including intent descriptions of the intent classification model from a language model, perform intent classification by using at least a portion of the dataset as an input prompt for the intent classification mode, and determine performance of the intent classification model by using at least a portion of the dataset as an input prompt for the intent classification model, wherein the dataset may include first data including independent intent descriptions in which intent names and intent descriptions for the intent names are mutually indistinguishable and second data including dependent intent descriptions in which intent names and intent descriptions for the intent names are mutually distinguishable.