Prompt-Guided Intention Recognition for Expanding Intent Types
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing user intention recognition systems in dialog-based human-computer interaction face challenges such as poor classification accuracy, the need for retraining when adding new intention types, and the requirement for super-large scale training, which affect their efficiency and reliability.
Innovation Solution
An intention recognition method utilizing a large model, including obtaining input sentences, generating first prompt information based on sentence-intention pairs, and performing intention recognition to determine the target intention, which enhances accuracy and reliability by limiting recognition within a predefined intention recognition range.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional classification methods are used for intention recognition, then the system structure is simple, but the classification accuracy is poor
Solution Approach 1:
The patent introduces prompt information as an intermediary element that bridges the input sentence and the classification process. The prompt information contains predefined intention types and classification criteria, acting as a mediator that guides the large model to achieve high classification accuracy without requiring complex system architecture changes
Solution Approach 2:
The patent changes the parameter of classification accuracy by utilizing large model technology with predefined prompt information. Instead of improving accuracy through complex system design, the approach changes the fundamental parameter by using pre-trained large models that inherently possess high classification capability when provided with appropriate prompt information
2Adaptability or versatility
If new intention types are added to the recognition system, then the system becomes more versatile, but retraining is required
Solution Approach 1:
The patent applies preliminary action by pre-defining multiple intention types and their corresponding classification criteria in the prompt information before actual use. When new intention types need to be added, they can be incorporated into the prompt information without requiring retraining of the large model, thus avoiding time loss while maintaining versatility
Solution Approach 2:
The large model serves as a universal classifier that can handle multiple intention types through a single unified architecture. By using prompt information to define different intention types, the system achieves multi-functionality without requiring separate classifiers or retraining for each new intention type, thereby maintaining adaptability while eliminating retraining time
3Measurement precision
If large scale training data is used, then the model becomes more accurate, but resource consumption increases
Solution Approach 1:
The patent applies preliminary action by using pre-trained large models that have already been trained on extensive data beforehand. This preliminary training is performed once, and the resulting model can be deployed for multiple recognition tasks without requiring additional large-scale training, thus achieving high accuracy while reducing ongoing resource consumption
Solution Approach 2:
Instead of training new models for each application, the patent uses copying by deploying the same pre-trained large model across different scenarios with different prompt information. This approach achieves high recognition accuracy through the pre-trained model's knowledge while avoiding the need to replicate training resources for each new application
4Adaptability or versatility
If the intention recognition range is expanded, then the system becomes more comprehensive, but reliability decreases
Solution Approach 1:
The patent applies segmentation by dividing the intention recognition space into distinct, predefined categories through prompt information. Each intention type is clearly defined with specific classification criteria, which segments the comprehensive recognition range into manageable, well-defined segments. This segmentation maintains reliability by providing clear decision boundaries while achieving comprehensiveness through multiple defined categories
Data Source
AI summary
An intention recognition method includes: obtaining an input sentence; obtaining at least one first sentence-intention pair based on the input sentence, in which the first sentence-intention pair includes a candidate sentence similar to the input sentence and a candidate intention to which the candidate sentence belongs; generating first prompt information based on the at least one first sentence-intention pair; and obtaining a target intention to which the input sentence belongs by performing an intention recognition on the input sentence based on the first prompt information.