Boundary detection for synthetic data generation

Patent metadata

Field	Value
Publication number	US-12399912-B1
Application number	US-202418762609-A
Country	US
Kind code	B1
Filing date	Jul 2, 2024
Priority date	Jul 2, 2024
Publication date	Aug 26, 2025
Grant date	Aug 26, 2025

Abstract

A computer-implemented method can determine a linguistic boundary condition for synthetic data generation in a multi-class classification problem. The method includes analyzing empirical labelled data using linguistic and vector representation techniques. The method further includes deriving a synthetic boundary conditional (SBC) model based on the analysis of the empirical labelled data and identifying a boundary location for performant synthetic data generation using the SBC model.

First claim

What is claimed is:

1. A computer-implemented method for determining a linguistic boundary condition for synthetic data generation in a multi-class classification problem, the method comprising:
analyzing empirical labelled data using linguistic and vector representation techniques;
deriving a synthetic boundary conditional (SBC) model based on the analyzing of the empirical labelled data; and
identifying a boundary location for performant synthetic data generation using the SBC model, comprising:
determining, by the SBC model, key collocations and key entities within each dataset class of the empirical labelled data; and
calculating 2D coordinates of the key collocations and the key entities.

2. The computer-implemented method of claim 1 , wherein the boundary location is defined by cartesian coordinates generated by the SBC model.

3. The computer-implemented method of claim 1 , wherein the analyzing of the empirical labelled data includes using corpus linguistics to determine key collocations of the empirical labelled data.

4. The computer-implemented method of claim 3 , wherein the analyzing of the empirical labelled data includes using named entity recognition to determine key entities within each label category.

5. The computer-implemented method of claim 4 , wherein the analyzing of the empirical labelled data includes representing key collocations and key entities in a multi-dimension space using word embeddings.

6. A computer-implemented method for determining a linguistic boundary condition for synthetic data generation in a multi-class classification problem, the method comprising:
analyzing empirical labelled data using linguistic and vector representation techniques;
deriving a synthetic boundary conditional (SBC) model based on the analyzing of the empirical labelled data, comprising:
performing corpus linguistic analysis to determine an order and arrangement of terms in the empirical labelled data;
performing name entity recognition (NER) to determine key entities of the empirical labelled data; and
mapping all words from multi-dimension word embeddings into a 2-dimensional cartesian space; and

identifying a boundary location for performant synthetic data generation using the SBC model.

7. The computer-implemented method of claim 1 , further comprising using the SBC model to measure a quality of synthetic data by calculating a similarity metric using the boundary location.

8. The computer-implemented method of claim 1 , further comprising using the SBC model to pass synthetic data requirements to a large language model for synthetic data generation.

9. A system comprising:
a processor;
a memory coupled to the processor; and
a computer readable storage embodying a computer program code, the computer program code comprising instructions for determining a linguistic boundary condition for synthetic data generation in a multi-class classification problem, the instructions executable by the processor and configured to:
analyze empirical labelled data using linguistic and vector representation techniques;
derive a synthetic boundary conditional (SBC) model based on the analyzing of the empirical labelled data; and
identify a boundary location for performant synthetic data generation using the SBC model,
wherein the instructions are further configured to derive the SBC model by:
performing corpus linguistic analysis to determine an order and arrangement of terms in the empirical labelled data;
performing name entity recognition (NER) to determine key entities of the empirical labelled data; and
mapping all words using word embeddings into a 2-dimensional cartesian space.

10. The system of claim 9 , wherein the boundary location is defined by cartesian coordinates generated by the SBC model.

11. The system of claim 9 , wherein the instructions are further configured to analyze the empirical labelled data by:
using corpus linguistics to determine key collocations of the empirical labelled data;
using named entity recognition to determine key entities within each label category; and
representing key collocations and key entities in a 2D cartesian space mapped from higher dimension word embeddings.

12. The system of claim 9 , wherein the instructions are further configured identify the boundary location using the SBC model by:
determining, by the SBC model, key collocations and key entities within each dataset class of the empirical labelled data; and
calculating 2D coordinates of the key collocations and the key entities.

13. The system of claim 9 , wherein the instructions are further configured to use the SBC model to measure a quality of synthetic data by calculating a similarity metric using the boundary location.

14. The system of claim 9 , wherein the instructions are further configured to use the SBC model to pass synthetic data requirements to a large language model for synthetic data generation.

15. A computer program product for determining a linguistic boundary condition for synthetic data generation in a multi-class classification problem, the computer program product comprising a computer readable storage medium having program instructions embodied therewith, the program instructions executable by a computer to cause the computer to:
analyze empirical labelled data using linguistic and vector representation techniques;
derive a synthetic boundary conditional (SBC) model based on the analyzing of the empirical labelled data; and
identify a boundary location, defined by 2D cartesian coordinates, for performant synthetic data generation using the SBC model by:
determining, by the SBC model, key collocations and key entities within each dataset class of the empirical labelled data; and
calculating the 2D coordinates of the key collocations and the key entities.

16. The computer program product of claim 15 , wherein the program instructions are further configured to analyze the empirical labelled data by:
using corpus linguistics to determine key collocations of the empirical labelled data;
using named entity recognition to determine key entities within each label category; and
representing key collocations and key entities in a 2D cartesian space mapped from higher dimension word embeddings.

17. The computer program product of claim 15 , wherein the program instructions are further configured to derive the SBC model by:
performing corpus linguistic analysis to determine an order and arrangement of terms in the empirical labelled data;
performing name entity recognition (NER) to determine key entities of the empirical labelled data; and
mapping all words using higher-dimension word embeddings into a 2-dimensional cartesian space.

Assignees

• IBM

Inventors

• Agrawal Bhavna
• Bhamidipaty Anuradha
• Sanjayam Jayshree
• Dunne Jonathan D
• Budhiraja Sarika

Classifications

• G06F16/285 Primary

  Clustering or classification · CPC title