Meeting Transcript Chunk Summarization for LLM Token Limits
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video conferencing systems face challenges in generating meeting minutes due to large transcript volumes exceeding the input token limits of large language models, leading to timeouts and inefficient processing when multiple users query the same transcript, causing concentration of calls at specific times and requiring high-performance systems.
Innovation Solution
A generative AI-based method that divides transcripts into data chunks, pre-summarizes them using a large language model, and stores the summary data, distributing the calls to the model over time, allowing for efficient generation and response to user queries.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If parallel processing is used to handle large transcripts, then processing capability is improved, but timeouts occur due to delays when transcript size exceeds certain limits
Solution Approach 1:
The patent divides the transcript into multiple data chunks based on configured units (time intervals or word counts). Each data chunk is processed independently through separate summarization calls to the large language model, avoiding the timeout problem associated with processing entire large transcripts at once. The system collects and merges the summary results from all data chunks to produce the final meeting minutes.
2Ease of operation
If meeting minutes are generated when users query about the meeting, then user-specific summaries are provided, but calls to the large language model concentrate at specific time zones causing duplicate operations
Solution Approach 1:
The patent performs preliminary summarization of data chunks as they are generated during the meeting, rather than waiting for user queries. The system continuously calls the large language model to summarize each data chunk when it becomes available, distributing the processing load over time. When users later query for meeting minutes, the pre-generated summary data is retrieved and merged, eliminating the need for concentrated LLM calls at query time.
3Loss of information
If the entire transcript is processed at once, then complete meeting minutes are generated, but the input token limit of the large language model is exceeded
Solution Approach 1:
The patent segments the transcript into multiple data chunks that each fit within the large language model's input token limits. The system processes each chunk independently through summarization, then merges the results to generate complete meeting minutes. This segmentation approach allows the system to handle transcripts of any size while respecting the token limits of the underlying language model.
Data Source
AI summary
The disclosure relates to a generative AI-based meeting minutes summarization method and apparatus. A generative AI (artificial intelligence)-based meeting minutes summarization method using a computing device according to an embodiment of the disclosure may include: collecting transcript texts generated during a meeting to generate data chunks for every configured unit; generating summary data, when the data chunks are generated, by summarizing the transcript texts included in the data chunks using a large language model (LLM); and generating, when a query is input, summarized meeting minutes by collecting the summary data generated up to the time at which the query is input.