Digital Assistant Reply Transparency via Generation Process Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Users face challenges in determining the credibility of digital assistant reply information and identifying the cause of inaccuracy, as the reference identifier provided does not sufficiently detail the generation process or specific content sources.
Innovation Solution
An information processing method that displays the generation process of reply information in an interactive window, allowing users to optimize segment content and improve accuracy by viewing segment matching results and modifying segment content directly.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If only reference identifier is provided in reply information, then the digital assistant can maintain simple operation interface, but the user cannot determine the credibility or identify the cause of inaccuracy in the reply information
Solution Approach 1:
The patent segments the generation process into distinct visual components: segment matching results showing which content segments are retrieved, reference identifiers linking to source documents, and inaccuracy location information pinpointing specific errors. This segmentation allows users to verify credibility through detailed breakdown without overwhelming the interface, as each component can be displayed selectively based on user interaction.
Solution Approach 2:
The patent adds a new dimension of transparency by displaying the generation process as a traceable pathway from query to reply. Instead of only showing the final reply, the system displays the intermediate steps (segment matching, reference selection, content generation) as visual evidence, enabling users to verify credibility by following this dimensional pathway from input to output.
2Manufacturing precision
If detailed generation process information is displayed to users, then users can identify and correct inaccuracies, but the interface complexity and information processing load increase
Solution Approach 1:
The patent performs preliminary actions by automatically generating and displaying the segment matching results, reference identifiers, and inaccuracy location information before the user needs to verify accuracy. This pre-computation and pre-display of verification information eliminates the need for users to manually trace through the generation process, maintaining ease of operation while enabling accuracy verification.
Solution Approach 2:
The patent introduces an intermediary layer between the digital assistant's internal generation process and the user interface. This intermediary automatically processes the complex generation information into user-friendly visual elements (highlighted segments, colored reference identifiers, annotated inaccuracy locations), shielding users from raw complexity while providing sufficient detail for accuracy verification.
3Productivity
If the digital assistant provides only reference identifiers, then the system can maintain simple structure, but users cannot optimize segment content to improve future reply accuracy
Solution Approach 1:
The patent implements feedback by displaying inaccuracy location information that directly points to problematic segments in the source content. Users can identify specific segments that caused errors and optimize them accordingly. This feedback loop enables continuous improvement of segment content quality, which directly enhances future reply accuracy without requiring complex retraining of the AI model.
Solution Approach 2:
The patent enables users to perform self-service optimization of segment content by providing them with the tools to identify and correct inaccuracies directly in the displayed generation process. Users can edit segment content based on the inaccuracy locations provided, and the system automatically incorporates these improvements into future responses, eliminating the need for complex administrative interfaces or developer intervention.
Data Source
AI summary
The present disclosure provides an information processing method, a device, and a medium. The method includes: displaying reply information of a digital assistant in an interactive window between a user and the digital assistant, the reply information is generated based on first request information of the user; and in response to a preset operation associated with the reply information, displaying first content, where the first content is content about a generation process of the reply information.