Response Phrase Selection for Ogiri Conversational Art
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies face difficulties in selecting appropriate response phrases for the 'ogiri' conversational art, as they struggle with low co-occurrence of response phrases with the theme phrase in large-scale text data and lack time-series elements, making it challenging to generate laughter-inducing responses.
Innovation Solution
A response phrase selection device that uses a positive/negative example determination model to calculate response phrase scores based on feature values, selecting a response phrase that matches the theme phrase by determining positive or negative examples, allowing for the selection of an appropriate response phrase.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a predicate-argument structure database is created from large-scale text data, then response phrases can be generated for common Q&A, but response phrases having low co-occurrence with ogiri theme phrases become positive examples making it difficult to construct a database that causes laughter
Solution Approach 1:
The patent segments the response phrase selection process into multiple independent modules: a positive/negative example determination model that classifies theme phrases, a feature value extraction system that identifies key characteristics, and a score calculation mechanism that ranks response phrases. This segmentation allows each module to specialize in one aspect of the problem, improving overall reliability for ogiri while maintaining versatility.
Solution Approach 2:
The patent introduces an intermediary classification layer that determines whether a theme phrase is a positive or negative example before selecting response phrases. This intermediary step acts as a mediator between the general text data and the specific ogiri requirements, filtering out inappropriate responses and ensuring that only laughter-inducing phrases are selected based on their co-occurrence patterns with theme phrases.
2Ease of operation
If a technique of scoring based on the history of words of spoken phrases of a user is used, then response phrases can be selected for ordinary conversation, but it is difficult to appropriately select a response phrase for ogiri which has few time-series elements
Solution Approach 1:
The patent implements a dynamic feature value extraction system that adapts to different conversation types. Instead of relying solely on fixed historical word sequences, the system dynamically identifies relevant features from both user history and theme phrases, adjusting the weighting and selection criteria based on whether the context is ordinary conversation or ogiri. This dynamic adaptation enables the system to handle ogiri's lack of time-series elements while maintaining ease of operation.
3Quantity of substance
If large-scale text data is used for response phrase generation, then common Q&A can be handled, but response phrases with low co-occurrence with theme phrases are selected reducing effectiveness for ogiri
Solution Approach 1:
The patent changes the evaluation parameters from simple co-occurrence frequency to a multi-dimensional scoring system that considers feature value similarity, positive/negative example classification, and thematic relevance. This parameter transformation allows the system to utilize large-scale text data effectively while improving matching accuracy for ogiri theme phrases, selecting responses that are both numerous and precisely aligned with the theme.
Data Source
AI summary
A theme phrase feature value calculation unit 103 calculates a feature value Fq of a given theme phrase Q based on the cosine similarity with a feature keyword of a feature keyword DB 107b. A positive/negative example determination unit 104 applies the feature value Fq to a positive/negative example determination model to determine whether a response phrase A is a positive example or a negative example. A response phrase score calculation unit 105 calculates the cosine similarity between the theme phrase Q and a plurality of theme phrases that form a pair with a response phrase A determined to be a positive example, and the highest value of the calculated cosine similarity is set as a score of the response phrase A. A response phrase selection unit 106 selects the response phrase indicating the maximum score as the response phrase for the theme phrase Q.