Triage Dialogue System Using Active Reward Learning for Task Completion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Intelligent triage systems face challenges in determining the completion of the triage task during human-machine interaction, leading to potential delays and inefficiencies in patient assessment, as existing methods do not explicitly evaluate task success and often require excessive interaction, wasting time and resources.
Innovation Solution
The implementation of an active reward learning method using a Gaussian process to build a reward model, which adjusts the triage dialogue policy to maximize the expected reward value, allowing for explicit determination of task completion and reducing the need for user feedback, thereby improving accuracy and efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the intelligent triage system interacts with the patient multiple times to collect comprehensive symptom and sign information, then the determination accuracy is improved, but the time consumption increases and patient visiting time is delayed
Solution Approach 1:
The patent implements a feedback mechanism where the system evaluates whether the collected symptom and sign information is sufficient to complete the triage task. Based on this evaluation feedback, the system dynamically determines whether to continue interacting with the patient or to proceed with the determination, thereby avoiding unnecessary interactions and time consumption while maintaining determination accuracy.
Solution Approach 2:
The system performs self-evaluation of task completion status by automatically assessing whether the collected information is sufficient. This self-service mechanism eliminates the need for manual judgment and enables the system to autonomously decide when to stop interaction, reducing time loss while ensuring accurate determination.
2Measurement precision
If the intelligent triage system interacts with the patient multiple times to collect comprehensive information, then the determination accuracy is improved, but the system complexity increases
Solution Approach 1:
The patent extracts the task completion evaluation function as a separate, independent module within the triage system. This extracted component specifically assesses whether collected information is sufficient, simplifying the overall system architecture by separating the evaluation logic from the main interaction and determination processes.
Solution Approach 2:
The system is segmented into distinct functional modules: information collection, task completion evaluation, and determination. This segmentation allows each module to operate independently with well-defined interfaces, reducing system complexity while maintaining the ability to achieve accurate determinations through coordinated module interactions.
3Ease of manufacture
If the traditional triage system uses manually written rule base or machine learning method, then the system implementation is simplified, but the task completion evaluation capability is lost
Solution Approach 1:
The patent merges the task completion evaluation capability with the existing rule-based or machine learning-based triage system. The evaluation module works in conjunction with the traditional determination methods, combining the simplicity of existing implementations with the added benefit of explicit task completion assessment to prevent information loss.
Data Source
Figure 1~2A
Figure 2B~3
Figure 4~5
AI summary
A triage dialogue method, a device and a system are provided. The triage dialogue method includes: receiving triage dialogue content (S202); dividing the triage dialogue content into a series of dialogue features, and generating dialogue feature vectors in one-to-one correspondence with the dialogue features (S2041); inputting the dialogue feature vectors into an encoder, and generating an intermediate representation vector through the encoder (S2042); determining whether the triage task is successfully completed according to the intermediate representation vector (S206'); and adjusting the triage dialogue policy according to the determination result of the triage task (S208).