Topic Selection Probability Adjustment for Dialogue Bias
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Interactive systems struggle to provide topics that users can easily discuss, as existing methods for extracting important words can be biased by the user's speech patterns, leading to biased topic selection and potential exclusion of topics the user finds difficult to talk about.
Innovation Solution
An interactive apparatus and method that acquires profile information associating topics with the number of specific parts of speech, sets higher selection probabilities for topics with larger counts of these parts of speech, and selects topics based on these probabilities to ensure the user can engage in dialogue about topics they are more comfortable discussing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If important words are extracted based on user speech frequency, then the system can identify topics the user talks about, but the topic selection becomes biased and excludes topics the user finds difficult to talk about
Solution Approach 1:
Instead of selecting topics based on what the user talks about frequently (conventional approach), the system inverts the logic by selecting topics based on what the user talks about least, thereby providing topics the user finds difficult to discuss and achieving more balanced topic coverage
Solution Approach 2:
The system changes the selection parameter from speech frequency to speech frequency complement (1 - frequency), transforming the selection criterion to prioritize underrepresented topics while maintaining quantitative basis for selection
2Ease of operation
If the system provides topics based on extracted important words, then it can respond to user interests, but it cannot actively provide topics about which the user can easily talk
Solution Approach 1:
The system performs preliminary analysis of user speech patterns to build a profile before topic selection, pre-calculating which topics are most suitable for the user based on their speech characteristics, thereby enabling active topic provision without complex real-time analysis
Solution Approach 2:
The system uses feedback from user speech data to continuously refine topic selection, analyzing the relationship between parts of speech and topics to improve topic recommendation accuracy over time while maintaining a relatively simple selection mechanism
Data Source
AI summary
An interactive apparatus includes: acquisition means for acquiring profile information in which a topic in a user's speech and the number of predetermined parts of speech of words included in the speech are associated with each other; topic selection means for setting the probability that the topic is selected in such a way that topics whose numbers of predetermined parts of speech are larger are selected with a higher probability from among topics set in the profile information based on the profile information acquired by the acquisition means, and selecting the topic in accordance with the probability of the selection that has been set; and response generation means for generating a response to the user based on the topic that has been selected by the topic selection means.