Voice-Adapted Answer Reformulation via Relevance Ranking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing voice-based user devices face limitations in scalability and accuracy when providing voice-based responses to queries, as they often rely on manual or simple automated processes that restrict the selection of answers to curated sources, reducing the ability to retrieve relevant information from a broader corpus of open-ended questions and answers.
Innovation Solution
A backend computer system that uses a machine learning model to reformulate answers by selecting relevant sentences from a larger corpus of open-ended questions and answers on the Internet, transforming non-presentable components into presentable ones, and ranking sentences based on relevance scores to generate concise, accurate responses.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual or simple automated processes are used to provide voice-based responses, then the system is easier to operate and implement, but the scalability and accuracy of retrieving relevant information from a broader corpus is reduced
Solution Approach 1:
The patent introduces a backend computer system as an intermediary between the voice-based user device and the corpus of questions and answers. This backend system performs the complex tasks of receiving audio data, determining questions, selecting answers, and generating reformulated answers, thereby resolving the contradiction by centralizing complexity in a dedicated intermediary system while keeping the user device simple
Solution Approach 2:
The system segments the overall processing into distinct phases: receiving audio data, determining the question, selecting an answer from multiple candidate answers, generating presentable sentences, and selecting the final reformulated answer. This segmentation allows each component to be optimized independently, improving accuracy while managing complexity through modular architecture
2Adaptability or versatility
If curated sources are used to limit answer selection, then the system is easier to manage and more reliable, but the ability to retrieve relevant information from a broader corpus is reduced
Solution Approach 1:
The system retrieves more candidate answers than are ultimately needed (excessive action), selecting from multiple candidate answers generated from the corpus before narrowing down to the final reformulated answer. This approach allows the system to explore a broader corpus for adaptability while maintaining reliability through selective filtering and ranking of the excessive candidates
Solution Approach 2:
The system employs feedback mechanisms where the backend computer system evaluates multiple candidate answers against the determined question, selects the most relevant ones, and uses this selection process to improve future answer generation. The feedback loop ensures that while the system can access a broad corpus, it maintains reliability by learning from and adapting to the quality of selected answers
3Ease of operation
If web-based answers are used directly without reformulation, then the process is simpler and faster, but the answers are not suitable for presentation via voice-based interface
Solution Approach 1:
The system performs preliminary actions by generating multiple presentable sentences from the selected answer before final selection. This preliminary transformation of the web-based answer into various sentence formats allows the system to simplify the final presentation step while having already done the complex transformation work in advance
Solution Approach 2:
The patent replaces manual or simple automated text processing with a machine learning model that automatically generates and selects presentable sentences. This substitution of mechanical processing with intelligent automation resolves the contradiction by handling the complexity of answer transformation through AI-driven processes while maintaining ease of operation for the end user
Data Source
AI summary
Techniques are disclosed. The techniques include receiving, by a computing device, an answer to a query, the answer comprising a content element and a metadata element. Based on the content element and the metadata element, the techniques generate a presentable sentence of a plurality of presentable sentences. The techniques then utilize a machine learning model to determine a relevance score for the presentable sentence based on the query and the presentable sentence, the relevance score being of a set of relevance scores and corresponding to a measure of relevance with respect to the presentable sentence answering the query. The techniques then select a portion of the plurality of presentable sentences based on a ranking of the set of relevance scores and transmit the portion to a user device for presentation on a voice-based interface of the user device.