Single-Turn Multi-Turn Request Classification for Voice Interfaces
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing voice recognition systems struggle to accurately understand user intentions and commands, particularly in multi-turn conversations, leading to repetitive confirmation requests that can cause user fatigue and dissatisfaction.
Innovation Solution
An electronic apparatus that identifies user requests as single-turn or multi-turn based on conversation content, generates confirmation requests only for single-turn requests, and uses a trained neural network model to estimate user preferences and reliability values to provide responses for multi-turn requests, thereby reducing repetitive confirmations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the system generates confirmation requests for all user requests to ensure accurate understanding, then the reliability of response is improved, but the user experiences fatigue and dissatisfaction due to repetitive confirmations
Solution Approach 1:
The patent segments user requests into two categories: single-turn requests and multi-turn requests. Single-turn requests are those that can be understood and responded to in one interaction, while multi-turn requests require multiple interactions. This segmentation allows the system to apply different confirmation strategies to different types of requests, generating confirmation requests only for single-turn requests where ambiguity is more likely, while avoiding repetitive confirmations for multi-turn requests where context already provides understanding.
2Measurement precision
If the system uses conversation history to improve understanding of user requests, then the accuracy of intent recognition is improved, but the system complexity increases
Solution Approach 1:
The patent applies preliminary action by pre-classifying user requests into single-turn and multi-turn categories before generating responses. The system uses the conversation history to predict whether a request is likely to be single-turn or multi-turn in advance, and based on this prediction, decides whether to generate a confirmation request. This preliminary classification simplifies the overall decision-making process compared to analyzing the entire conversation history for every request, reducing computational complexity while maintaining accurate intent recognition.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
An electronic apparatus may include a communication interface, a memory configured to store at least one instruction, and one or more processors configured to generate response information based on conversation content received through the communication interface by executing the at least one instruction to: identify a user request by confirming a current request and a previous request of a user in the conversation content, identify whether the user request is a multi-turn request or a single-turn request based on the conversation content, generate a confirmation request to request the user to confirm the user request based on the user request being identified as the single-turn request, and omit a generation of the confirmation request and generate a response corresponding to the user request based on the user request being identified as the multi-turn request.