Triage Dialogue System Using Active Reward Learning for Task Completion

Resolve Bottlenecks,
Find Innovative Solutions
Generate 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

VSEngineering 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

Engineering Contradiction:
Improvedetermination accuracyVSAvoidpatient visiting time
Core Design Contradiction:
Measurement precisionVSLoss of time

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.

Inventive Principle:
Principle #23Feedback

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.

Inventive Principle:
Principle #25Self-service

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

Engineering Contradiction:
Improvedetermination accuracyVSAvoidsystem complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

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.

Inventive Principle:
Principle #2Taking out (Extraction)

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.

Inventive Principle:
Principle #1Segmentation

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

Engineering Contradiction:
Improvesystem implementationVSAvoidtask completion evaluation capability
Core Design Contradiction:
Ease of manufactureVSLoss of information

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.

Inventive Principle:
Principle #5Merging (Combining)

Data Source

PatentEP3660854B1Triage dialogue method, device, and system
Publication Date: 2023.08.16 BOE TECHNOLOGY GROUP CO LTD
  • EP3660854B1 patent drawingFigure 1~2A
  • EP3660854B1 patent drawingFigure 2B~3
  • EP3660854B1 patent drawingFigure 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).