Head-Mounted Assistant Response Stitching for Multi-Intent Queries
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems face challenges in integrating different information perspectives to generate a multi-perspective response that is coherent and natural-sounding, determining which execution results to stitch together, and establishing the correct order for stitching these results.
Innovation Solution
The assistant system determines dialog-intents associated with a user query, selects agents accordingly, and uses relevance scores, filtering out mutually exclusive results, and employs a sequential-language model, predefined rules, user profile data, and linguistic grounding to stitch execution results in a logical and personal manner.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If the system waits for all execution results before generating a response, then the completeness of information is improved, but the response time increases
Solution Approach 1:
The system performs preliminary actions by identifying and presenting the most relevant execution results first, based on relevance scores, before all results are available. This allows the system to provide timely partial responses while continuing to process remaining results in the background.
Solution Approach 2:
The system applies partial action by selecting and presenting only the top-k most relevant execution results rather than waiting for all results. This partial presentation achieves a balance between response speed and information completeness, allowing the system to respond quickly with the most important information.
2Loss of information
If the system integrates multiple information perspectives, then the informativeness of the response is improved, but the complexity of integrating results increases
Solution Approach 1:
The system segments the integration process by first ranking individual execution results based on relevance scores, then selectively combining only the top-k results. This segmentation approach simplifies the integration complexity by breaking down the overall task into ranking and selective combination steps.
Solution Approach 2:
The system changes the parameter of result selection from binary (include/exclude) to graded (relevance-based ranking). By introducing relevance scores as a parameter, the system can dynamically determine which results to include, making the integration process more flexible and manageable.
3Loss of information
If the system presents all execution results, then the completeness of information is improved, but the naturalness of the response decreases
Solution Approach 1:
The system changes the parameter of result presentation from exhaustive to selective by introducing relevance scores. This allows the system to present only the most relevant results in a natural, conversational manner, avoiding the robotic presentation of all available information.
Solution Approach 2:
The system applies partial action by presenting only the top-k most relevant execution results rather than all results. This partial presentation maintains naturalness by focusing on the most important information while still providing sufficient completeness for effective user interaction.
Data Source
AI summary
In one embodiment, a method includes receiving a user query inputted on a head-mounted device from the head-mounted device, wherein the user query corresponds to multiple dialog-intents, executing multiple tasks corresponding to the multiple dialog-intents, generating a multi-perspective response by a stitching model based on two or more of execution results of the multiple tasks, wherein the stitching model combines the two or more of the execution results based on natural language processing, and wherein the multi-perspective response comprises a natural-language response combining the two or more execution results, and sending instructions to the head-mounted device for presenting the multi-perspective response on the head-mounted device.