Situation-Based Dialogue Response Selection for Multi-Task Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Dialogue systems requiring expertise, such as expert systems, incur high construction costs due to the need for manual scenario creation, and combining multiple tasks further increases costs.
Innovation Solution
A dialogue apparatus that stores examples with situation information, uses a selection rule to select appropriate responses based on dialogue states, and includes a voice recognition and synthesis unit to facilitate low-cost construction and execution of multiple tasks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a scenario method is used to construct a dialogue system for achieving a predetermined task, then the dialogue system can execute tasks requiring expertise, but the construction cost becomes very large due to manual scenario creation by experts
Solution Approach 1:
The patent uses example data from social networking services as templates or copies of real user interactions. Instead of creating scenarios from scratch by experts, the system collects and stores actual dialogue examples that naturally occur on social media platforms. These examples are then used as the basis for constructing the dialogue system, significantly reducing the manual effort and cost required while maintaining task execution capability.
Solution Approach 2:
The system automatically collects example data from social networking services without requiring manual scenario creation by experts. The dialogue examples are gathered automatically from real user interactions on social media platforms, and the system self-organizes this data into usable formats. This automated data collection and processing eliminates the need for expensive manual scenario development while preserving the expertise embedded in real user dialogues.
2Adaptability or versatility
If scenarios created by multiple experts are combined to construct a dialogue system that executes multiple tasks, then the system can handle various tasks, but more cost is required compared to constructing a single-task system
Solution Approach 1:
The patent creates a universal dialogue system that can handle multiple tasks by collecting diverse example data from social networking services. Instead of creating separate scenarios for each task by different experts, the system gathers a broad range of dialogue examples that naturally cover multiple topics and tasks. This single unified system inherits multi-task capability from the diverse examples, eliminating the need to combine multiple expert-created scenarios and reducing overall construction costs.
3Ease of operation
If an example method is used in chat dialogue systems with simple utterance-response rules, then the system can engage in casual dialogue, but it cannot achieve accurate dialogue flows for tasks requiring expertise
Solution Approach 1:
The patent performs preliminary organization and categorization of example data before use. Example data collected from social networking services is pre-processed, stored in structured formats, and organized by situation types and dialogue patterns. This preliminary preparation ensures that when the system needs to respond, it can quickly retrieve appropriate examples rather than searching through raw data, thereby maintaining simple operation while achieving accurate task-oriented dialogue flows.
Solution Approach 2:
The system uses situation information as feedback to select appropriate examples. When a user input is received, the system determines the current situation type and uses this feedback to retrieve examples that match the situation. This feedback mechanism ensures that the simple example-based system can accurately select relevant responses for task-oriented dialogues, bridging the gap between simplicity and accuracy.
Data Source
AI summary
An example storage unit (10-1) stores examples including an utterance sentence, a response sentence, and situation information. A selection rule storage unit (10-3) stores a selection rule including a dialogue state, available situation information, and a dialogue state of a transition destination. An utterance reception unit (12) receives a user utterance uttered. A dialogue state acquisition unit (13) acquires a current dialogue state. An example selection unit (14) selects a selection example in which the situation information corresponds to situation information available in the current dialogue state and the utterance sentence corresponds to the user utterance, from the examples, by using the selection rule. A dialogue state update unit (15) updates the current dialogue state to the dialogue state of the transition destination included in the selection rule. An utterance presentation unit (16) presents a system utterance based on a response sentence included in the selection example.