One-Shot Document Data Capture Using Anchor-Based Kernel Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Extracting information from semi-structured electronic documents is challenging due to their lack of well-defined layouts, making it difficult to reliably parse and extract relevant data.
Innovation Solution
The One-Shot Data Capture (OSDC) technique uses a labelled document to determine anchors with an unlabelled document, generating a kernel to identify corresponding elements using a kernel machine, such as SVM, RKDA, or GP, to extract key-value pairs and table data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional parsing methods are used on semi-structured documents, then the processing speed is maintained, but the extraction accuracy deteriorates due to lack of well-defined layouts
Solution Approach 1:
The patent uses a labelled document as a template or copy to guide the extraction process from the unlabelled document. The labelled document contains pre-defined structures and annotations that are copied over to identify corresponding elements in the target document, enabling accurate extraction without complex parsing logic.
Solution Approach 2:
The patent introduces a labelled document as an intermediary between the extraction system and the unlabelled document. This intermediary provides the structural guidance and semantic annotations needed to accurately identify and extract relevant information from semi-structured documents without requiring complex document analysis algorithms.
2Measurement precision
If one-shot data capture with kernel machines is used, then data extraction accuracy is improved, but computational complexity increases
Solution Approach 1:
The patent performs preliminary actions by pre-defining the labelled document with all necessary annotations, structures, and semantic information before the actual extraction process. This preliminary preparation enables the kernel machine to work with pre-processed data, reducing the computational burden during the actual extraction phase while maintaining high accuracy.
3Measurement precision
If anchors are determined between labelled and unlabelled documents, then alignment accuracy is improved, but processing time increases
Solution Approach 1:
The patent segments the document alignment process by identifying and matching specific anchor points (such as headers, footers, or key structural elements) between the labelled and unlabelled documents. Rather than attempting to align every element, the system focuses on critical anchor points that define the document structure, thereby achieving accurate alignment with reduced processing time.
Data Source
AI summary
There is provided a method (100) for performing data capture in an unlabelled document based on a labelled document, wherein the labelled document contains one or more elements each labelled with respect to a field. The method comprises: determining (S110) one or more anchors, wherein an anchor comprises a pair of points that match between the labelled document and the unlabelled document; generating (S120) an overall kernel using the determined one or more anchors; and identifying (S130), using the overall kernel in a kernel machine, one or more elements to be captured in the unlabelled document, wherein the one or more elements to be captured correspond in field to the one or more labelled elements contained in the labelled document.