Response Sentence Generation Using Speech Type Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for generating response sentences in chat dialog systems require large amounts of data and struggle to adapt to different user characters, especially in social networking services where various character types are present, leading to inconsistent responses.
Innovation Solution
A response sentence generating device and method that analyzes input speech sentences using a text analyzing unit, determines speech types, and employs a type conversion rule to generate appropriate response types from a database, allowing for the generation of response sentences without the need for extensive data preparation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a method based on large amount of data is used to generate response sentences, then the response sentences can be generated with high accuracy, but the data preparation becomes complex and time-consuming
Solution Approach 1:
The patent extracts only the essential features from speech sentences (speech type classification) rather than using the complete raw data. The text analyzing unit extracts key linguistic features, and the response-type determining unit classifies speech types based on these extracted features, eliminating the need for large amounts of raw training data while maintaining generation accuracy.
Solution Approach 2:
The patent segments the speech sentence analysis into distinct functional components: text analyzing unit for feature extraction, response-type determining unit for classification, and response-sentence generating unit for generation. This segmentation allows each component to process only necessary information independently, reducing overall data requirements.
2Measurement precision
If a method based on large amount of data is used to generate response sentences, then the response sentences can be generated with high accuracy, but the device complexity increases
Solution Approach 1:
The patent removes the complex data preparation stage by extracting only essential speech features directly from input sentences. The text analyzing unit and response-type determining unit work together to classify speech types without requiring pre-prepared large datasets, thereby reducing device complexity while maintaining accuracy.
Solution Approach 2:
The patent introduces an intermediary classification layer (response-type determining unit) that mediates between the input speech sentence and the response generation process. This intermediary unit classifies speech types based on extracted features, simplifying the overall system architecture by adding a single functional layer rather than requiring complex data processing pipelines.
3Adaptability or versatility
If speeches from multiple different characters are included, then the system can handle diverse characters, but the speeches become inconsistent
Solution Approach 1:
The patent applies local quality by determining response types specifically for each speech type category rather than using a unified approach for all characters. The response-type determining unit analyzes the local characteristics of each speech type (e.g., emotional speech, informative speech) and determines appropriate response types accordingly, allowing consistent responses within each category while maintaining character adaptability.
Solution Approach 2:
The patent segments the diverse character speeches into distinct speech type categories through the text analyzing unit and response-type determining unit. By classifying speeches into specific types (emotional, informative, etc.), the system can handle multiple characters consistently within each category, resolving the inconsistency problem while maintaining versatility.
Data Source
AI summary
To make it possible to generate a response sentence with respect to an input speech sentence without preparing a large amount of data.A response-type determining unit 117 determines, based on an analysis result of a speech sentence analyzed by a speech-content analyzing unit 112, a speech type indicating a type of the speech sentence and determines a response type with respect to the determined speech type based on the speech type and a type conversion rule prescribing, for each speech type, a rule for a response type with respect to a speech of the speech type. A response-sentence generating unit 119 generates the response sentence based on the response type and a response sentence database.