Multi-granularity Meeting Summarization via Event Ranker
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current speech-to-text technologies produce transcripts that include semantically insignificant filler words and lack user-customizable summarization options, making them difficult to read and consume.
Innovation Solution
A customizable dialogue summarization system that allows users to control summarization granularity, readability, speaker focus, and topic selection through a user interface, utilizing a re-trained language model and event ranker to generate summaries.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If speech-to-text technology provides a complete transcript of all utterances, then the record is faithful and complete, but the transcript includes filler words and becomes difficult to read
Solution Approach 1:
The patent segments the transcript into multiple granularities including sentence level, paragraph level, and section level summaries. This allows the system to maintain complete records at the sentence level while providing condensed summaries at higher levels, resolving the contradiction between completeness and readability.
Solution Approach 2:
The system dynamically adjusts the level of summarization based on user preferences and context. Users can select different granularity levels (e.g., executive summary, detailed summary, or full transcript), allowing the system to adapt between providing complete records and readable summaries depending on needs.
2Loss of information
If the transcript includes every utterance with filler words, then the record is complete, but the content becomes excessive and hard to consume
Solution Approach 1:
The patent applies partial summarization by selectively condensing portions of the transcript rather than processing the entire text uniformly. The system identifies and summarizes specific sections (introductions, conclusions, key points) while maintaining detailed records of important content, reducing overall volume while preserving essential information.
Solution Approach 2:
Different parts of the transcript are treated with different levels of summarization quality. Critical sections maintain higher detail and fidelity, while less critical portions receive more aggressive summarization. This local differentiation reduces overall text volume while preserving information quality where needed.
3Device complexity
If the summarization is fixed and generic, then the system is simple to implement, but it cannot be customized to user preferences
Solution Approach 1:
The system implements dynamic configurability where summarization parameters (granularity level, focus areas, length) can be adjusted based on user preferences, document type, and context. This allows the same system to adapt to different user needs without requiring multiple separate systems.
Solution Approach 2:
The patent creates a universal summarization framework that can handle multiple summarization styles and granularities through a single system. The core summarization engine serves multiple functions by adjusting parameters, eliminating the need for separate simple and complex systems.
4Loss of information
If the summary is too long and detailed, then it covers all topics thoroughly, but it loses the benefit of being a concise summary
Solution Approach 1:
The patent segments summaries into hierarchical levels: executive summaries provide high-level topic coverage in brief form, while detailed summaries and full transcripts are available for deeper exploration. This segmentation allows users to quickly assess topic coverage at the executive level and drill down only when needed.
Solution Approach 2:
The system performs preliminary summarization at multiple levels before user consumption. Executive summaries are generated first to provide quick topic coverage, allowing users to determine if they need to invest time in reading more detailed versions. This preliminary action filters content based on user information needs.
Data Source
AI summary
Generally discussed herein are devices, systems, and methods for. A method can include receiving, from a user through a user interface, a segmentation granularity value indicating a number of events in the transcript to be included in a summary, extracting, by a ranker model and from the transcript, a number of hints equal to the number of events, generating, by a summarizer model that includes a re-trained language model, respective summaries, one for each event, of a portion of the transcript corresponding to the event, and providing the respective summaries as an overall summary of the transcript.