Vector Generation for Table Data Captions

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing methods for natural language processing, such as the bag of words (BoW) method, cannot handle table data with captions, and techniques like low rank matrix completion and collaborative filtering struggle to incorporate text descriptions, limiting machine learning and prediction capabilities.

Innovation Solution

A data processing device and method that generates vectors by combining the frequency of word appearances in row and column captions of table data cells, enabling machine learning and prediction on table data.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If the bag of words method is used for natural language processing, then text and linguistic expressions can be vectorized and processed, but table data with captions cannot be handled

Engineering Contradiction:

Improvecapability to handle different data typesVSAvoidmethod limitations

Core Design Contradiction:

Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent extends the bag of words method to handle both text data and table data with captions by treating table captions as natural language descriptions. The same vectorization approach is applied universally to both text and table captions, enabling a single method to process multiple data types without requiring separate specialized techniques for each type.

Inventive Principle:

Principle #6Universality (Multi-functionality)

2Adaptability or versatility

If techniques like low rank matrix completion and collaborative filtering are used to handle table data, then table data can be processed, but text descriptions in natural language cannot be incorporated

Engineering Contradiction:

Improvecapability to incorporate text and table dataVSAvoidfeature extraction limitations

Core Design Contradiction:

Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent merges the processing of table data and text data by applying the bag of words vectorization method to table captions. This combines the strengths of table data processing with natural language processing capabilities, allowing both types of data to be handled using the same vectorization framework and enabling machine learning models to process both table and text information simultaneously.

Inventive Principle:

Principle #5Merging (Combining)

3Loss of information

If table data is processed using existing methods, then table structure can be maintained, but natural language descriptions in captions are lost

Engineering Contradiction:

Improvepreservation of natural language informationVSAvoidmachine learning and prediction capability

Core Design Contradiction:

Loss of informationVSProductivity

Solution Approach 1:

The patent extracts the natural language captions from table data and applies vectorization to them using the bag of words method. By extracting and processing the caption information separately as natural language descriptions, the method preserves the semantic meaning of the table data while enabling machine learning models to process and utilize the natural language information for predictions.

Inventive Principle:

Principle #2Taking out (Extraction)

Data Source

PatentUS11900051B2Data processing device, data processing method, and data processing program

Publication Date: 2024.02.13 NIPPON TELEGRAPH & TELEPHONE CORP

AI summary

A data processing device (10) includes a vector generation unit (1411, 1421) configured to obtain a number of appearance or a frequency of appearance of a word used in a caption of a row of a cell and in a caption of a column of the cell at which the row and the column of table data intersect with each other and generate a vector which has the obtained number of appearance or the obtained frequency of appearance as a value of an element.