Document Content Block Comparison via Graph Neural Network
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Comparing contents of two document files with multiple versions and varying revisions is challenging due to differences in component arrangement, making it difficult to identify corresponding components across different versions.
Innovation Solution
A method involving an electronic device that preprocesses content blocks in document files to extract properties such as text, geometric, and visual properties, and then uses a graph neural network (GNN) structure to compare these properties across content blocks in different documents, generating a comparison result.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If traditional component-based comparison methods are used to identify corresponding components in document files, then the comparison can be performed systematically, but the accuracy deteriorates when component arrangements differ significantly between document versions
Solution Approach 1:
The patent segments document content into discrete content blocks with defined properties (text, geometric, visual). Each content block is independently analyzed and represented as a node in a graph structure, allowing systematic comparison while maintaining the ability to handle arrangement variations through property-based matching rather than position-based matching.
Solution Approach 2:
The patent transforms the comparison problem from spatial arrangement-based to property-based by extracting and comparing multiple properties (text content, geometric properties like position and size, visual properties like font and formatting) of content blocks. This parameter transformation enables accurate correspondence identification even when spatial arrangements differ between document versions.
2Stability of the object's composition
If document comparison focuses on maintaining consistent component arrangement, then layout fidelity is preserved, but the ability to detect actual content differences deteriorates when revisions involve reorganization
Solution Approach 1:
The patent applies different comparison strategies to different properties of content blocks. Text properties are compared for semantic equivalence, geometric properties capture spatial relationships, and visual properties preserve formatting information. This local quality differentiation allows the system to detect content differences while accounting for legitimate reorganization in document revisions.
Solution Approach 2:
The patent uses a dynamic graph neural network approach where the graph structure and node properties can adapt to represent different document versions. The comparison process dynamically identifies correspondences based on property similarity rather than fixed positional relationships, enabling detection of content changes even when layouts are reorganized.
3Measurement precision
If detailed property extraction is performed on each content block to enable accurate comparison, then comparison precision is improved, but computational complexity increases
Solution Approach 1:
The patent employs a unified graph neural network framework that handles multiple property types (text, geometric, visual) through a single comparative mechanism. This universal approach simplifies the processing system by providing a multi-functional comparison engine that can handle diverse property types without requiring separate complex processing pipelines for each property category.
Solution Approach 2:
The patent introduces a graph structure as an intermediary representation between raw document content and comparison results. The graph nodes encapsulate content block properties and relationships, serving as a mediator that organizes detailed property information in a structured manner that facilitates efficient comparison while managing computational complexity through hierarchical processing.
Data Source
AI summary
A method for comparing content of two document files each having a plurality of content blocks is provided. The method is to be implemented by an electronic device and includes the steps of: performing, for the each of the content blocks in each of the document files, a pre-process operation so as to obtain a plurality of properties associated with the content block; comparing, for each content block from one of the document files, the properties thereof with the properties of each of the plurality of content blocks of the other one of the document files; and generating a comparison result based on the operations of the comparing.