Conversation Controller Plan Segmentation for Unclear Input
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional conversation controllers struggle to provide coherent and relevant responses to user utterances that are unclear or lack coherence, leading to ineffective counseling in situations where expert knowledge is needed.
Innovation Solution
A conversation controller with a plan memory unit and a plan conversation unit that divides explanatory sentences into multiple reply sentences, allowing for sequential output based on user input and next candidate designation information, enabling the controller to provide coherent responses even when user utterances are unclear or unrelated.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If the conventional conversation controller outputs answers completely dependent on user utterance, then the response is directly tied to user input, but the controller cannot understand unclear utterances and output suitable replies
Solution Approach 1:
The system performs preliminary actions by storing multiple candidate reply sentences and their associated confidence scores in advance. When a user utterance is unclear, the system can select from pre-prepared candidates rather than attempting to generate a response from scratch, thereby maintaining reliability even when understanding accuracy is low.
Solution Approach 2:
The system changes the parameter of response generation by introducing confidence score thresholds. When the confidence score falls below a predetermined threshold, the system switches from generating direct responses to selecting from candidate replies, thus adapting to unclear utterances while maintaining response quality.
2Ease of operation
If the conversation controller divides explanatory sentences into multiple reply sentences, then coherent responses can be provided sequentially, but the system complexity increases with plan memory and candidate designation
Solution Approach 1:
The system segments explanatory sentences into multiple reply sentences, each stored as a separate plan with associated confidence scores. This segmentation allows the controller to output coherent responses sequentially while managing complexity through structured storage and confidence-based selection rather than complex generative processes.
Solution Approach 2:
The system uses confidence scores as feedback mechanisms to determine whether to output candidate replies or seek clarification. This feedback loop simplifies the control structure by automatically deciding when to use pre-prepared candidates versus when to attempt direct response generation, reducing the need for complex manual control logic.
3Adaptability or versatility
If the conversation controller withholds and resumes explanations based on user interest, then conversation relevance is improved, but the control logic becomes more complex
Solution Approach 1:
The system uses confidence scores as feedback to automatically determine when to withhold or resume explanations. When confidence is high, the system withholds candidate replies and provides direct explanations; when confidence is low, it resumes using candidate replies. This feedback mechanism simplifies control logic by replacing complex judgment about user interest with automatic confidence-based decisions.
Solution Approach 2:
The system changes the parameter of response generation dynamically based on confidence score thresholds. By adjusting the threshold parameter, the system can adapt its behavior to be more or less responsive to user interest without changing the underlying control structure, thus improving adaptability while keeping control logic relatively simple.
Data Source
AI summary
A conversation controller outputs a reply sentence according to a user utterance. The conversation controller comprises a conversation database and a conversation control unit. The conversation database stores a plurality of plans. Each plan has a reply sentence and one or more pieces of next plan designation information for designating a next candidate reply sentence to be output following the reply sentence. The conversation control unit selects one of the plans stored in the conversation database according to a user utterance and outputs a reply sentence which the selected plan has. Then, the conversation control unit selects one piece of the next plan designation information which the plan has according to a next user utterance and outputs a next candidate reply sentence on the basis of the selected piece of the next plan designation information. Some plans have a plurality of reply sentences into which one explanatory sentence is divided.