Tabular Data Parsing via Row and Vertical Clustering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing document parsing technologies fail to effectively organize and extract information from arbitrarily formatted documents, such as inspection reports, where information is not logically partitioned due to varying organizational structures.
Innovation Solution
A method for parsing tabular data in documents by clustering rows based on proximity and numeric content, followed by generating vertical clusters within these row clusters, allowing for the identification and extraction of specific information using pattern matching techniques.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional parsing methods are used on arbitrarily formatted documents, then the parsing process is simple, but the information organization becomes erroneous and illogical
Solution Approach 1:
The patent segments the document into discrete row elements and clusters them based on spatial proximity and numeric content similarity. This segmentation approach allows the system to handle arbitrarily formatted documents by breaking them into manageable units that can be independently analyzed and reorganized logically, resolving the contradiction between simple parsing and accurate information organization.
Solution Approach 2:
The patent changes the parameters used for clustering from traditional spatial-only metrics to a combination of spatial proximity and numeric content similarity. This parameter transformation enables the system to maintain reliability in information organization while managing complexity through standardized clustering algorithms that operate on these transformed parameters.
2Reliability
If rows are clustered based on proximity and numeric content, then information organization improves, but processing time increases
Solution Approach 1:
The patent applies partial clustering by focusing on numeric content similarity as a key criterion, rather than analyzing all possible document attributes. This selective approach maintains high clustering accuracy for tabular data while reducing processing time by avoiding exhaustive analysis of non-numeric elements.
Solution Approach 2:
The patent transforms the clustering problem by introducing numeric content similarity as a new parameter alongside spatial proximity. This parameter change enables more accurate clustering of tabular data while the system manages processing time through efficient calculation of numeric similarity metrics.
3Measurement precision
If vertical clusters are generated within row clusters, then information extraction accuracy improves, but system complexity increases
Solution Approach 1:
The patent segments the clustered rows into vertical clusters based on columnar patterns and alignment. This segmentation enables precise extraction of tabular information by identifying vertical relationships among rows, while the modular nature of the segmentation process keeps system complexity manageable through reusable clustering algorithms.
Solution Approach 2:
The patent transitions from horizontal row-based clustering to vertical column-based clustering by analyzing character alignment and pattern matching across rows. This dimensional transformation improves information extraction precision by capturing the vertical structure of tabular data while managing complexity through pattern recognition techniques.
Data Source
AI summary
One or more techniques and/or systems are provided for parsing tabular data of a document. That is, a document may comprise arbitrarily formatted content (e.g., an equipment inspection report generated by an engineer). Respective rows of the document may be clustered into one or more row clusters based upon row proximity and/or numeric content (e.g., rows having similar numeric content may comprise logically related information). One or more vertical clusters may be generated within respective row clusters based upon vertical overlap. In this way, row clusters and/or vertical clusters may be searched for one or more values that may be assigned to a search term. For example, a row cluster may comprise a search term “Average temp”. One or more vertical clusters within the row cluster may be searched for a word that matches a pattern criteria (e.g., a two digit number), which may be assigned to the search term.