Information Extraction Using Visual Layout and Markup Rules
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing rule-based information extraction frameworks fail to effectively utilize visual cues in web page layouts, leading to limitations such as sensitivity to minor modifications, complex specification languages, and errors in markup code, as they rely on source code translations and are not robust to different implementations of visual patterns.
Innovation Solution
A method combining spatial and textual layout cues for information extraction, using visual layout rules, markup rules, and text-based rules to extract information from formatted documents, with visual operator algebra for efficient retrieval and storage of visual regions, enabling robustness to layout variations and errors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If existing rule-based information extraction frameworks use source code patterns, then they can extract information using markup rules, but they fail to utilize visual cues and are sensitive to different implementations of the same visual pattern
Solution Approach 1:
The patent transitions from analyzing one-dimensional source code sequences to analyzing two-dimensional visual layouts by rendering the document and detecting spatial relationships between elements. This dimensional change enables the system to capture visual proximity cues that are lost in linear source code representation, making extraction rules robust to different markup implementations of the same visual pattern.
Solution Approach 2:
The patent introduces a visual layout analysis intermediary that sits between the document source code and the information extraction rules. This intermediary renders the document and extracts visual layout information, converting it into a form that can be used by extraction rules. This mediator enables the system to work with visual patterns independently of their specific markup implementation.
2Ease of operation
If existing approaches rely on HTML tags and DOM trees, then they can process structured data, but they become sensitive to minor modifications and rule maintenance becomes messy
Solution Approach 1:
The patent creates a visual copy or representation of the document layout by rendering it and extracting spatial relationships. Instead of working directly with the fragile DOM tree structure, the system works with a visual representation that captures the essential layout information. This copying approach makes the extraction rules insensitive to minor markup modifications while maintaining ease of operation.
3Measurement precision
If text-based analysis is used on specification languages, then information can be extracted from markup code, but the analysis becomes complex and errors in markup code reduce accuracy
Solution Approach 1:
The patent replaces complex text-based analysis of specification languages with a visual analysis approach. Instead of parsing and analyzing markup code textually, the system renders the document and analyzes visual spatial relationships. This substitution simplifies the analysis process while improving accuracy by leveraging the browser's rendering capability to handle markup errors automatically.
Data Source
AI summary
Techniques for extracting information from a formatted document are provided. The techniques include combining one or more visual layout rules, one or more mark-up rules and one or more text-based rules in connection with a formatted document, and specifying one or more rules from the one or more visual layout rules, one or more mark-up rules and one or more text based rules to extract information from the formatted document.