Document Structure Estimation Using Sentence-Bound Tree Pruning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for estimating document structure in unstructured documents often incorrectly detect list elements, leading to mistakes such as mingling of foreign elements with in-line lists or omission of elements within the same sentence, due to the lack of a common structural definition and varied document formats.
Innovation Solution
A method that extracts candidate elements from documents, groups them based on extraction rules and adjacent element classifications, builds trees with root and leaf nodes, and prunes trees to identify unbranched structures, ensuring elements are correctly grouped within a single sentence, preventing incorrect detections and omissions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional techniques are used to estimate document structure, then processing speed is maintained, but detection accuracy deteriorates with frequent mistakes in identifying list elements
Solution Approach 1:
The patent segments the document processing into distinct stages: extracting candidate elements, grouping them by sentence boundaries, building trees within each group, and pruning inconsistent structures. This segmentation allows accurate detection by handling one sentence group at a time, preventing the mingling of elements from different sentences while maintaining manageable processing complexity through systematic decomposition
Solution Approach 2:
The patent performs preliminary grouping of candidate elements by sentence boundaries before building trees. This preliminary action ensures that elements are pre-organized into sentence-specific groups, preventing foreign elements from different sentences from being incorrectly included in the same list structure, thereby improving detection accuracy before the actual tree construction occurs
2Reliability
If conventional techniques detect list elements based on fixed format, then processing simplicity is maintained, but reliability deteriorates due to varied document structures
Solution Approach 1:
The patent implements dynamic tree building where the structure is not predetermined but emerges from the actual relationships among candidate elements within each sentence group. The tree construction adapts to the specific document structure by identifying parent-child relationships based on element positioning and logical connections, allowing the system to handle varied document formats reliably while maintaining reasonable processing complexity through algorithmic flexibility
Solution Approach 2:
The patent employs feedback mechanisms through the tree pruning stage, where initially built trees are evaluated and inconsistent structures are removed. This feedback loop refines the detected document structure by comparing against expected structural patterns and eliminating erroneous detections, thereby improving reliability while the systematic feedback process keeps processing complexity manageable
3Measurement precision
If elements are grouped without sentence boundary consideration, then processing efficiency is maintained, but measurement precision deteriorates with frequent omission of elements
Solution Approach 1:
The patent segments candidate elements into sentence-specific groups before tree construction, ensuring that each group contains only elements from a single sentence. This segmentation improves element identification accuracy by preventing the omission of elements that should be included in the same sentence's list structure, while processing efficiency is maintained by applying the same tree-building algorithm to each segmented group independently
Solution Approach 2:
The patent performs preliminary grouping of candidate elements by sentence boundaries before the main tree construction process. This preliminary action ensures complete element identification by establishing sentence-specific boundaries that prevent omission of relevant elements, while processing efficiency is preserved by organizing elements into manageable groups that can be processed systematically through the tree building and pruning stages
Data Source
AI summary
A system and method for estimating document structure of a document which includes extracting one or more candidate elements describing the document structure from the document and grouping the one or more candidate elements into a group and building one or more trees for the group. Each tree has a root node and a leaf node selected from the candidate elements in the group. The method further includes pruning the one or more trees while leaving a path from the root node to the leaf node, based on whether a text corresponding to the path to the leaf node is accommodated in a single group of words.