Hierarchical Document Classification System Reducing Processing Time
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Document data processing is time-consuming and prone to errors, especially when manual, and existing automated systems require large datasets and immense processing power, making them inefficient and costly.
Innovation Solution
A hierarchical document classification system (HDCS) that uses machine learning classifiers to classify documents based on image and textual data, dynamically selecting between image-based and text-based classifiers to minimize memory requirements and processing power, allowing for efficient classification with a smaller training dataset.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If manual document processing is used, then flexibility and adaptability are maintained, but processing time increases and error rates rise
Solution Approach 1:
The patent segments the document processing task into multiple hierarchical levels (first level classification, second level classification, third level classification). Each level handles specific aspects of document categorization, allowing parallel processing of different document aspects and reducing overall processing time while maintaining accuracy through specialized classifiers at each level.
Solution Approach 2:
The system dynamically changes parameters such as selecting different classification models (e.g., CNN, LSTM, Transformer) based on document characteristics. It adjusts processing depth and model complexity according to the specific document type, optimizing the balance between processing speed and accuracy for different scenarios.
2Productivity
If automated classification systems are used, then processing speed increases, but computing power requirements and memory needs increase significantly
Solution Approach 1:
The patent divides the classification system into hierarchical levels where simpler models handle initial filtering and more complex models handle specific document types. This segmentation allows the system to use computationally efficient models for the majority of documents while reserving heavy computational resources only for complex cases, reducing overall energy consumption.
Solution Approach 2:
The system applies partial classification action by performing only the necessary level of classification for each document. Not all documents require full hierarchical classification - many are accurately classified at earlier levels, avoiding unnecessary computational overhead and memory usage for complete classification of every document.
3Measurement precision
If deep learning models are used for document classification, then accuracy improves, but training data requirements and computational resources increase
Solution Approach 1:
The patent segments the training process into multiple stages corresponding to different hierarchical levels. Each level is trained on progressively more specialized data, allowing the system to achieve high accuracy with smaller, more focused training datasets at each stage rather than requiring one massive dataset for all classification tasks.
Solution Approach 2:
The system performs preliminary classification at upper hierarchical levels before applying more complex models. This preliminary action pre-filters documents, so that expensive deep learning models only process a subset of documents, effectively reducing the amount of training data needed for those complex models while maintaining overall accuracy.
Data Source
AI summary
A method performed by a computing system comprises receiving, by a hierarchical document classification system (HDCS), a document that includes image data and textual data, wherein the document is associated with a document type. The HDCS determines a first machine learning (ML) classifier of the HDCS associated with the document type. The first ML classifier corresponds to one of an image-based ML classifier of the HDCS and a text-based ML classifier of the HDCS. The determined first ML classifier classifies the document as being associated with a particular class of document of a plurality of classes of documents. A second ML classifier associated with the class of document classifies the document as being associated with a particular subclass of a plurality of subclasses of the class of document. The HDCS communicates information that specifies the class and the subclass associated with the document to a terminal.