Text Block Masking for Document Classification Efficiency
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning language model classification systems face efficiency and reliability shortcomings in document classification, particularly in handling long documents and providing informative explanations of predictions.
Innovation Solution
The method involves generating unmasked label probability scores and segmenting documents into text blocks, then performing classification iterations with masking to generate masked label probability scores. These scores are used to calculate text block importance scores, ultimately producing a predictive data output that identifies influential text blocks for classification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If traditional machine learning classification systems process entire documents at once, then classification can be performed, but computational operations and training data requirements increase significantly
Solution Approach 1:
The patent divides a document into multiple text blocks and processes them separately through the classification model. Instead of feeding the entire document at once, the system segments the input into manageable chunks (e.g., paragraphs or sections), classifies each block independently, and then aggregates the results. This segmentation reduces the computational load per inference while maintaining classification effectiveness.
2Measurement precision
If text block masking is performed to assess influence, then predictive accuracy improves, but processing time increases
Solution Approach 1:
The patent applies text block masking selectively rather than exhaustively. Instead of masking all possible text blocks across all documents uniformly, the system identifies and masks only those text blocks that are most relevant or influential to the classification decision. This partial application of masking reduces the overall processing time while still capturing the essential information needed for accurate predictions.
3Reliability
If all text blocks are processed and masked to generate importance scores, then classification reliability improves, but the amount of training data required increases
Solution Approach 1:
The patent assigns different levels of processing quality to different text blocks based on their local characteristics. Text blocks that are more influential or contain more critical information receive more intensive processing (including masking and importance score generation), while less critical blocks receive lighter processing. This localized quality approach ensures that the most important portions of the document contribute most to classification reliability, while reducing the overall training data requirements.
Data Source
AI summary
Various embodiments of the present disclosure provide methods, apparatus, systems, computing devices, computing entities, and/or the like for processing document classification system outputs, wherein classification routine iterations are performed using masked document data objects comprising one or more masked text blocks. Text block importance score for text blocks are generated and compared to generate predictive data output comprising text blocks determined to be the most influential in classifying the document data objects with respect to one or more classification labels.