Multi-model Neural Network for Unstructured Document Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing DCA systems face challenges in efficiently processing unstructured and semi-structured electronic information, as they struggle to translate this data into numerical representations without relying on resource-intensive brute force algorithms.
Innovation Solution
A novel multi-model, multi-task trained neural network architecture is introduced, which breaks down the analysis task into auxiliary tasks and uses multiple sub-models to parse and label different electronic format types within documents, enabling efficient feature extraction and content analysis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If brute force algorithms are used to translate unstructured and semi-structured electronic information into numerical representations, then the translation can be performed, but the computing resources required become excessively large
Solution Approach 1:
The patent segments the complex task of translating unstructured electronic information into numerical representations by introducing a document layout model that divides documents into distinct format types (text, tables, figures, lists). Each format type is then processed by specialized machine learning models, avoiding the need for brute force algorithms to handle all formats uniformly. This segmentation enables efficient processing while maintaining reliable translation capability.
2Productivity
If multiple specialized machine learning models are used to process different electronic format types, then processing efficiency improves, but the system complexity increases
Solution Approach 1:
The patent implements a universal document layout model that serves multiple functions: it parses documents, identifies format types, and structures output for different model types. This multi-functional approach allows the system to handle various electronic formats (text, tables, figures, lists) through a unified architecture, improving processing efficiency while controlling system complexity through shared infrastructure.
Solution Approach 2:
The document layout model acts as an intermediary between the raw electronic document and the specialized machine learning models. It translates diverse document formats into a standardized structure that multiple models can process, enabling efficient multi-format handling without requiring each model to independently parse raw documents, thus managing system complexity.
3Device complexity
If a single machine learning model is used to process all electronic format types, then the system complexity remains low, but the ability to accurately process unstructured and semi-structured information deteriorates
Solution Approach 1:
The patent applies local quality by assigning specialized machine learning models to specific electronic format types (text models for text, table models for tables, figure models for figures, list models for lists). Each model is optimized for its designated format type, achieving high processing accuracy for unstructured and semi-structured information while maintaining manageable system complexity through the universal document layout model that orchestrates these specialized components.
Data Source
AI summary
Embodiments of the invention describe a computer-implemented method of analyzing an electronic version of a document. The computer-implemented method can include an architecture of machine learning sub-models that performs the global task of translating unstructured and semi-structured inputs into numerical representations that can be recognized and manipulated by a content-analysis (CA) sub-model without relying on brute force analysis. Embodiments of the invention achieve these results by separating the global task into auxiliary tasks and assigning each sub-model to at least one of the auxiliary tasks. The auxiliary tasks can include parsing the unstructured or semi-structured inputs into format types (e.g., lists, tables, figures, text, etc. of a PDF document), extracting features of the parsed document, and performing a computer-based CA on the extracted features. The sub-models are trained in stages and in groups, wherein both the stages and the groupings are based on the complexity of the sub-model's assigned task.