Multi-Sentence Characteristic Extraction With Context-Window Ensembles
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies struggle to efficiently analyze and extract multiple-sentence characteristics from large volumes of technical documents required for storage device development, particularly in the context of solid state drives (SSDs), due to the complexity and variability of these documents.
Innovation Solution
A system and method utilizing a plurality of multi-label classifiers and an ensemble neural network to analyze technical documents, employing different context window sizes for improved accuracy in extracting multiple-sentence characteristics, leveraging large language models to generate multi-labels for sentences.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple classifiers with different context window sizes are used to analyze technical documents, then the accuracy of extracting multiple-sentence characteristics is improved, but the device complexity and processing time increase
Solution Approach 1:
The patent divides the classification task into multiple specialized classifiers, each handling different context window sizes (e.g., 1-sentence, 2-sentence, 3-sentence contexts). This segmentation allows each classifier to focus on specific characteristic extraction patterns, improving overall accuracy while maintaining manageable complexity through modular design
Solution Approach 2:
The patent introduces the dimension of context window size variation, transforming a single-classifier approach into a multi-classifier ensemble that processes the same text at different contextual depths. This dimensional expansion enables capturing both local sentence-level features and broader paragraph-level characteristics simultaneously
2Measurement precision
If multiple classifiers with different context window sizes are used to analyze technical documents, then the accuracy of extracting multiple-sentence characteristics is improved, but the processing time increases
Solution Approach 1:
The patent performs preliminary processing by dividing the document into sentences and organizing them into overlapping context windows before classification. This pre-organization allows multiple classifiers to work in parallel on pre-prepared inputs, reducing the overall processing time compared to sequential analysis
Solution Approach 2:
The patent merges the results from multiple classifiers into a unified characteristic extraction output. By combining predictions from classifiers with different context window sizes, the system achieves high accuracy while optimizing processing efficiency through result integration rather than reprocessing
3Adaptability or versatility
If classifiers process text fragments with different context window sizes, then the ability to capture various sentence characteristics is improved, but the difficulty of detecting and measuring characteristics increases
Solution Approach 1:
The patent designs classifiers that are universal in their ability to handle different context window sizes and detect various sentence characteristics (requirements, constraints, specifications). Each classifier is multi-functional, capable of identifying multiple characteristic types within its context window, reducing the overall difficulty of detection
Data Source
AI summary
A system for analyzing technical documents for storage devices and extracting multiple-sentence characteristics. The system includes: a plurality of classifiers, each classifier configured to receive multiple sentences from the technical document and generate multi-labels for the multiple sentences, each label indicating whether each sentence has a target characteristic described in the technical document; and an ensemble neural network configured to sequentially receive, as training datasets, multiple multi-labels from the plurality of classifiers, and generate, as a result of training, multiple labels for the multiple sentences based on the training datasets. Each of the plurality of classifiers is configured to receive text fragments at different datapoints corresponding to the multiple sentences with different context window sizes, and generate the multi-labels corresponding to the text fragments.