Nested LLM Summarization of Large Conversation Records
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing summary tools are limited in their ability to effectively analyze and summarize large sets of text data, particularly in communication systems, and lack the capability to generate comprehensive summaries from multiple independent text pieces with varying relationships.
Innovation Solution
A computer-implemented method using a large language model (LLM) to process and summarize large sets of text data by employing chunking algorithms, constructing language responses, and aggregating summaries to generate individual and aggregated summaries of communication history records.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If existing summary tools are used to analyze large sets of text data, then the summary generation process is simple, but the capability to effectively analyze and summarize large sets of text data is limited
Solution Approach 1:
The patent divides the large set of text data into multiple independent text pieces using a chunking algorithm. Each text piece is processed separately through the LLM to generate individual summaries. These summaries are then aggregated to produce a comprehensive summary of the entire dataset. This segmentation enables the system to handle large datasets effectively while maintaining manageable processing complexity through structured division of tasks.
2Productivity
If all text data is processed individually, then the summary quality is high, but the processing resources and time consumed are excessive
Solution Approach 1:
Instead of processing every single text piece individually (excessive action), the patent uses a sampling approach where a subset of text pieces is selected and processed. The LLM generates summaries for these sampled text pieces, and then aggregates these summaries to represent the entire dataset. This partial action significantly reduces processing time and resource consumption while still producing useful summary insights, accepting that not every individual text piece is processed in detail.
3Adaptability or versatility
If a single LLM call is used to summarize all data, then the process is simple, but the ability to handle varying relationships between text pieces is limited
Solution Approach 1:
The patent implements a nested summarization structure where individual text pieces are first summarized to create intermediate summaries, and then these intermediate summaries are aggregated to create a final comprehensive summary. This nested approach allows the system to handle varying relationships between text pieces by processing them at different levels of abstraction. The chunking algorithm ensures that text pieces with similar relationships are grouped together, enabling the system to capture complex relationships while maintaining a manageable multi-step processing structure.
Data Source
AI summary
A device may receive, at a server system, a free-form query. The device may select two-way communication history records from a database associated with the free-form query to identify selected history records. The device may generate individual summaries of individual history records of the selected history records by processing a set of corresponding individual history records using a chunking algorithm, constructing language responses from outputs of the chunking algorithm using a large language model, and aggregating the language responses. The device may process the language responses using the large language model to generate an individual summary for corresponding individual history records.