Knowledge Graph Table Processing Retaining Row Column Relationships

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing methods for information extraction from complex tables in documents often lose layout information, leading to reduced accuracy in understanding unstructured content due to simple extraction of value fields without retaining row and column relationships.

Innovation Solution

A method and apparatus for processing data based on a knowledge graph that acquires and recognizes table content, determines row and column attributes using a preset table lexicon, and generates a quadruple list to retain contextual information, ensuring accurate extraction and retention of table relationships.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If simple extraction of value fields is performed, then extraction speed is improved, but layout information is lost leading to reduced accuracy

Engineering Contradiction:
Improveextraction speedVSAvoidaccuracy of understanding unstructured content
Core Design Contradiction:
ProductivityVSMeasurement precision

Solution Approach 1:

The patent segments table extraction into multiple processing stages: initial value field extraction, row attribute identification, column attribute identification, and quadruple list generation. This segmentation allows parallel processing of different table components while maintaining their relationships, thus improving both speed and accuracy.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent transforms traditional two-dimensional table data into a four-dimensional quadruple structure (table_name, row_attribute, column_attribute, attribute_value). This dimensional expansion preserves layout information by explicitly encoding row and column relationships, enabling accurate reconstruction of table semantics while maintaining extraction efficiency.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

2Loss of information

If row and column relationships are retained, then integrity of extracted content is improved, but processing complexity increases

Engineering Contradiction:
Improveintegrity of extracted contentVSAvoidprocessing complexity
Core Design Contradiction:
Loss of informationVSDevice complexity

Solution Approach 1:

The patent extracts row attributes and column attributes as separate identifiable entities from the table structure. By isolating these structural elements and representing them as distinct quadruples, the system maintains table integrity without requiring complex processing of the entire table structure at once, thus reducing processing complexity while preserving information完整性.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The quadruple list serves as an intermediary data structure that bridges the gap between raw table data and semantic understanding. Each quadruple encapsulates a complete semantic unit with its layout context, acting as a mediator that preserves row-column relationships without requiring complex ongoing processing, thereby maintaining integrity while simplifying subsequent operations.

Inventive Principle:
Principle #24Intermediary (Mediator)

3Measurement precision

If attribute identification based on lexicon matching is performed, then extraction accuracy is improved, but processing time increases

Engineering Contradiction:
Improveaccuracy of attribute identificationVSAvoidprocessing time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent performs preliminary segmentation of table content into candidate row attributes and column attributes before full lexicon matching. This preliminary classification narrows down the search space for subsequent detailed matching, allowing accurate attribute identification through lexicon comparison while significantly reducing the overall processing time by avoiding exhaustive matching of all table cells.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP4113328A1Method and apparatus for processing data based on knowledge graph, electronic device and medium
Publication Date: 2023.01.04 BEIJING BAIDU NETCOM SCI & TECH CO LTD
  • EP4113328A1 patent drawingFigure 1~2
  • EP4113328A1 patent drawingFigure 3~4
  • EP4113328A1 patent drawingFigure 5

AI summary

Disclosed are a method for processing data, an electronic device and a medium. The technical solution includes: acquiring (101, 301) a table to be processed and a corresponding table name; recognizing (102, 302) the table to acquire each cell content in the table; determining (103) a row attribute and a column attribute corresponding to each cell contents based on a matching degree between each cell content and a word segmentation in a preset table lexicon; and determining (104, 305) a quadruple list corresponding to the table based on the table name, the row attribute and the column attribute corresponding to each cell content.