Neural Network Document Classification via Positional Context
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing document classification systems rely heavily on text-based models and require large amounts of training data, making them inefficient and error-prone, especially when dealing with scanned images of documents.
Innovation Solution
A unified document classification system that uses a neural network architecture to extract data from scanned images by performing optical character recognition, semantic enrichment, and positional context analysis, allowing for efficient classification with a limited set of training data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If OCR technology is used to process text from scanned images, then text extraction is achieved, but all data including irrelevant data is produced leading to classification inefficiency
Solution Approach 1:
The patent extracts only the relevant terms of interest from the OCR-produced text data, rather than processing all extracted text. This is achieved by identifying and isolating specific keywords or phrases that are pertinent to document classification, thereby eliminating irrelevant data and improving classification efficiency while maintaining complete text extraction capability
Solution Approach 2:
The patent segments the document image into multiple regions or zones and processes each region separately to identify terms of interest. This segmentation approach allows the system to focus on specific areas containing relevant information, reducing the overall data processing burden while ensuring comprehensive text extraction from the entire document
2Ease of operation
If keyword search is performed without contextual correlation, then simple text matching is achieved, but document classification accuracy deteriorates
Solution Approach 1:
The patent introduces an intermediary processing step between keyword extraction and classification that establishes contextual relationships. This intermediary layer analyzes the spatial and semantic context of extracted terms, correlating them with their positions and relationships in the original document, thereby maintaining processing simplicity while significantly improving classification accuracy
Solution Approach 2:
The patent adds a contextual dimension to the keyword search process by considering not only the presence of keywords but also their positional context, surrounding text, and relationship to other elements in the document. This multi-dimensional approach transforms simple text matching into context-aware classification while preserving operational simplicity
3Reliability
If conventional machine learning techniques are used for document extraction, then data processing is achieved, but thousands of example documents are required as training data
Solution Approach 1:
The patent performs preliminary processing and enrichment of the training data before model training, including semantic enrichment and structural organization of documents. This preliminary action prepares the data in a more effective format, allowing the model to learn from fewer examples by maximizing the information content and quality of each training document, thereby reducing the total quantity needed while maintaining robustness
4Measurement precision
If human intervention is used to determine relevant text from OCR data, then accurate text selection is achieved, but labor intensity and error probability increase
Solution Approach 1:
The patent implements a self-service system where the neural network automatically identifies and selects relevant terms from OCR data without human intervention. The system uses trained models to autonomously determine text relevance, perform semantic enrichment, and classify documents, thereby maintaining high accuracy while achieving complete automation and eliminating human error in the selection process
Data Source
AI summary
A system to classify image of a document using neural network architecture is provided. The system includes a storage device storing the image derived from the document having text information. The system includes a document importer operable to perform optical character recognition to convert image data in the image to machine readable data. The system includes a neural network that perform semantic enrichment and positional context for the terms of interest present in the image. The neural network is configured to take as input the machine-readable data and the image and combine both the machine-readable data and the image to classify the image of the document based on the positional context of the terms of interest.