Transformer Model Document Classification via Text Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies face challenges in accurately classifying documents based on their text content, particularly in distinguishing between different types of documents such as receipts, credit card statements, and missing receipt affidavits.
Innovation Solution
A system that utilizes a transformer-based machine learning model to determine the type of document by analyzing the text extracted from document images. The model is trained using a set of training data that includes text from various documents and their corresponding labels, allowing it to generate probabilities for different document types and select the most likely type.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional machine learning models are used for document classification, then the system is simpler to implement, but the classification accuracy is lower
Solution Approach 1:
The patent transforms the document classification problem from image-based to text-based by changing the input parameter format. Text extraction converts document images into textual data, which is then processed by machine learning models to predict document types. This parameter transformation enables the use of more accurate text-processing models while maintaining reasonable system complexity.
2Reliability
If text extraction from document images is performed, then the input quality for machine learning models improves, but the processing time increases
Solution Approach 1:
The system performs text extraction as a preliminary action before feeding data into the machine learning model. By pre-processing the document images to extract text content in advance, the system improves the quality of input data for classification while separating the extraction phase from the prediction phase, allowing for optimized processing workflows.
Data Source
AI summary
Some embodiments provide a non-transitory machine-readable medium that stores a program. The program receives an image of a document, the document comprising a set of text. The program further provides the set of text to a machine learning model configured to determine, based on the set of text, a plurality of probabilities for a plurality of defined types of documents. Based on the plurality of probabilities for the plurality of defined types of documents, the program also determines a type of the document from the plurality of defined types of documents.