Automated Table Annotation via Unstructured Text Matching

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

The manual process of annotating and labeling datasets for machine-learning models is time-consuming and prone to human error, especially when dealing with tables in non-native representations, such as PDF documents, where table elements are not easily identifiable by computers.

Innovation Solution

A system and method that automatically annotate table elements by matching text elements from unstructured table representations with table elements from structured representations, using techniques like similarity measures and fuzzy matching to generate annotated datasets, reducing the need for manual labeling.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If manual annotation of table elements is performed, then annotation accuracy can be achieved, but time consumption and labor cost increase significantly

Engineering Contradiction:

Improveannotation accuracyVSAvoidtime consumption

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The system creates a copy of the unstructured table representation and uses it as a reference to automatically annotate the structured table representation. By copying text elements from the unstructured version and matching them with elements in the structured version, the system achieves accurate annotation without manual intervention for each element.

Inventive Principle:

Principle #26Copying

Solution Approach 2:

The manual mechanical process of annotating each table element is replaced with an automated computational system. The system uses text element identification, matching algorithms, and automatic tag generation to substitute human annotators, thereby maintaining accuracy while dramatically reducing time consumption.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

2Adaptability or versatility

If tables in non-native representations (e.g., PDF) are processed, then broader document compatibility is achieved, but table element identifiability deteriorates

Engineering Contradiction:

Improvedocument compatibilityVSAvoidtable element identifiability

Core Design Contradiction:

Adaptability or versatilityVSDifficulty of detecting and measuring

Solution Approach 1:

The system introduces an intermediary unstructured table representation that serves as a bridge between the non-native PDF format and the structured table format. This intermediary contains text elements that can be identified and matched, thereby enabling the processing of PDF tables while maintaining element identifiability through the matching process.

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The system performs preliminary extraction of text elements from the unstructured table representation before the matching process. By pre-identifying and storing text elements with their positions, the system prepares the data in advance, making it easier to match with structured table elements and overcome the identifiability issues of non-native representations.

Inventive Principle:

Principle #10Preliminary action

3Productivity

If automated annotation is implemented, then productivity increases, but annotation precision may deteriorate due to matching errors

Engineering Contradiction:

Improveannotation speedVSAvoidannotation precision

Core Design Contradiction:

ProductivityVSMeasurement precision

Solution Approach 1:

The system incorporates feedback mechanisms in the matching process by comparing text elements from the unstructured representation with elements in the structured representation. The matching algorithm receives feedback from similarity measures and adjusts its matching decisions accordingly, allowing automated processing while maintaining precision through iterative refinement.

Inventive Principle:

Principle #23Feedback

Solution Approach 2:

The system changes parameters such as similarity thresholds and matching criteria to optimize the balance between speed and precision. By adjusting these parameters, the system can achieve high productivity while maintaining annotation precision, resolving the contradiction between automated speed and accuracy.

Inventive Principle:

Principle #35Parameter changes

Data Source

PatentUS11551146B2Automated non-native table representation annotation for machine-learning models

Publication Date: 2023.01.10 INTERNATIONAL BUSINESS MACHINE CORPORATION

AI summary

One embodiment provides a method, including: receiving two documents, one of the two documents having at least one table that includes the same information as a corresponding table in the other of the two documents, wherein (i) one of the two documents comprises the at least one table in an unstructured table representation and (ii) the other of the two documents comprises the at least one table in a structured table representation; identifying text elements within the at least one table in the unstructured table representation; matching the identified text elements with table elements within the at least one table in the structured table representation; and annotating the at least one table in the structured table representation based upon the matches between the table elements and text elements.