Text Report Concept Generation via Clustering and Predictive Modeling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Automatic text summarization using sophisticated models is impractical due to high processing power requirements and often generates duplicative or unsuitable summaries, especially when dealing with large volumes of user feedback reports.
Innovation Solution
A system that automatically generates concepts from large text corpuses by partitioning reports into clusters using a clustering subsystem, selecting representative text, and training predictive models with augmented datasets to reduce computational resources and labor, while accurately identifying repeated concepts.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If sophisticated text summarization models (deep neural networks) are used, then the quality of text summarization is improved, but the processing time and computational resources increase dramatically
Solution Approach 1:
The patent segments the summarization task into two distinct stages: (1) extractive summarization that selects and ranks important sentences from the input text, and (2) abstractive summarization that generates final summaries only for clustered groups of similar reports. This segmentation allows the system to process individual reports quickly using lightweight extractive models, then apply more computationally intensive abstractive models only to representative summaries of clustered reports, dramatically reducing overall processing time while maintaining quality.
2Measurement precision
If sophisticated text summarization models are used, then the quality of text summarization is improved, but the computational resources required increase dramatically
Solution Approach 1:
The patent merges multiple similar reports into clusters before applying resource-intensive abstractive summarization. By identifying reports with similar topics, sentiments, or key entities and grouping them together, the system generates a single abstractive summary for the entire cluster rather than processing each report individually. This merging strategy maintains high summarization quality for diverse report types while dramatically reducing computational resource consumption by eliminating redundant processing of similar reports.
3Productivity
If automatic text summarization is applied to large volumes of reports, then the coverage of processed reports is improved, but the results become highly duplicative
Solution Approach 1:
The patent performs preliminary clustering and similarity analysis on reports before generating abstractive summaries. By pre-grouping reports into clusters based on their semantic similarity, topic overlap, or sentiment alignment, the system identifies which reports are duplicates or near-duplicates. This preliminary action enables the system to skip generating separate summaries for redundant reports and instead generate a single representative summary for the entire cluster, maintaining high information value while eliminating duplicative content.
4Speed
If text summarization is applied to short reports (around 100 words), then the processing speed is improved, but the output simply repeats the input
Solution Approach 1:
The patent introduces an intermediary extractive summarization stage that acts as a bridge between short input reports and abstractive summary generation. For reports around 100 words or shorter, the extractive model first identifies and selects the most important sentences or phrases, creating a condensed intermediate representation. This intermediary step provides the abstractive model with focused, high-value content to work with, enabling it to generate meaningful summaries even from very short inputs without simply repeating the original text.
Data Source
AI summary
Methods, systems, and apparatus, including computer programs encoded on computer storage media, for automatic text report concept generation. Generating concepts from text reports includes receiving a collection of text reports; performing a clustering process for a plurality of different cluster sizes; evaluating each of the plurality of different cluster sizes to select an optimal cluster size; generating, from the collection of text reports, clusters using the selected optimal cluster size; aggregating text associated with text reports in each cluster; maintaining a training dataset comprising the aggregated text; and generating a predictive model from the training dataset to generate a concept for an input text report.