Counter Utterance Generation Model Using Argumentative Schemes
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Dialogue systems face challenges in generating appropriate responses to input utterances that deviate from pre-defined argumentative topics, and existing utterance generation models struggle to control output utterances according to argumentation methodologies, leading to performance declines and increased data collection costs.
Innovation Solution
A generation apparatus and method that includes an argumentative scheme adding unit, a generation model learning unit, and a counter utterance generating unit, which learns a generation model using argumentative scheme-added pair data to generate counter utterances based on input utterances and designated argumentative schemes, employing techniques like Conditional Variational Auto Encoders to control output utterances.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manually collecting pair data constituted by input utterance and output utterance for each methodology of argumentation is adopted, then control of output utterance according to argumentation methodology is improved, but data collection cost increases and performance declines due to division of learning data
Solution Approach 1:
The patent merges multiple argumentation methodologies into a single unified learning model. Instead of creating separate models for each argumentation type (rebuttal, concession, etc.), the system collects pair data across all argumentation types and trains one comprehensive utterance generation model, thereby reducing data collection costs while maintaining control precision through the unified framework
Solution Approach 2:
The utterance generation model is designed with multi-functionality to handle various argumentation methodologies simultaneously. The model learns to generate appropriate output utterances for different argumentation types (rebuttal, concession, support, etc.) within a single framework, eliminating the need for separate specialized models and reducing overall system complexity
2Measurement precision
If manually collecting pair data constituted by input utterance and output utterance for each methodology of argumentation is adopted, then control of output utterance according to argumentation methodology is improved, but performance declines due to division of learning data
Solution Approach 1:
The patent combines learning data from multiple argumentation methodologies into a unified dataset for training a single model. This merging approach prevents the performance degradation that occurs when data is divided across multiple separate models, as the unified model benefits from the combined data volume and can learn generalizable patterns across different argumentation types
3Adaptability or versatility
If creating graph data for discussing arbitrary topics is adopted, then adaptability to different topics is improved, but implementation becomes unrealistic due to infinite number of argumentative topics
Solution Approach 1:
The system employs self-service through automatic pair data generation using dialogue history and argumentation knowledge bases. Instead of requiring manual creation of graph data for each topic, the system automatically generates training data by analyzing dialogue contexts and retrieving relevant argumentation patterns, making the system scalable to arbitrary topics without proportional increases in manual effort
Solution Approach 2:
The patent performs preliminary action by pre-collecting and organizing argumentation knowledge and dialogue history before actual topic discussion. The system prepares training data in advance by gathering relevant argumentation patterns and dialogue examples, which can then be efficiently applied to various topics without requiring real-time manual graph creation
Data Source
AI summary
A generation apparatus 100 includes: an argumentative scheme adding unit 10 which adds an argumentative scheme with respect to pair data constituted by an input utterance and a counter utterance 121 that voices a negative opinion with respect to the input utterance and which generates argumentative scheme-added pair data 122; a generation model learning unit 20 which learns a generation model for generating a counter utterance from an input utterance in consideration of the argumentative scheme by using the argumentative scheme-added pair data 122 as learning data and which generates a learned counter utterance generation model 123; and a counter utterance generating unit 30 which acquires an input utterance of a user and a designated argumentative scheme and which outputs a counter utterance using the counter utterance generation model 123.