Retrieval-based, self-supervised augmentation using transformer models

Patent metadata

Field	Value
Publication number	US-12360977-B2
Application number	US-202318191896-A
Country	US
Kind code	B2
Filing date	Mar 29, 2023
Priority date	Mar 29, 2023
Publication date	Jul 15, 2025
Grant date	Jul 15, 2025

Abstract

Embodiments of the invention are directed to a computer system comprising a memory communicatively coupled to a processor system. The processor system is operable to perform processor system operations that include accessing query information associated with a to-be-augmented information set (TBAIS) having a TBAIS format. Query information sequence vectors (QISV) are generated that represent the query information and the TBAIS. Unannotated data repository information sequence vectors (UDRSV) are accessed that represent unannotated data repository information having a plurality of information formats. Matching UDRSV are identified, where the matching UDRSV include the UDRSV that match the QISV. A response to the query information is generated based at least in part on the matching UDRSV.

First claim

What is claimed is:

1. A computer system comprising a processor system coupled to a memory, wherein the processor system is configured to perform processor system operations comprising:
accessing query information associated with a to-be-augmented information set (TBAIS) having a TBAIS format;
applying the query information, the TBAIS, and the TBAIS format to a sequencer to generate a sequence;
applying the sequence to a non-parametric transformer of the processor system to generate query information sequence vectors (QISV) that represent the query information and the TBAIS;
accessing unannotated data repository information sequence vectors (UDRSV) that represent unannotated data repository information having a plurality of information formats;
applying an approximate nearest neighbor (ANN) search technique of the processor system to identify matching UDRSV comprising the UDRSV that match the QISV;
wherein the non-parametric transformer and the ANN search technique provide a traceable path to how the ANN search technique identified the matching UDRSV; and
generating a response to the query information based at least in part on the matching UDRSV.

2. The computer system of claim 1 , wherein the processor system operations further comprise incorporating the response into the TBAIS to generate an augmented version of the TBAIS.

3. The computer system of claim 1 , wherein the TBAIS format is different from at least one of the plurality of information formats.

4. The computer system of claim 3 , wherein the TBAIS format is selected from the group consisting of a database, a table, and a portable document format (PDF).

5. The computer system of claim 3 , wherein the information format is selected from the group consisting of a database, a table, and a portable document format (PDF).

6. The computer system of claim 1 , wherein the processor system operations further comprise using the traceable path to generate an explanation of how the ANN search technique identified the matching UDRSV.

7. The computer system of claim 1 , wherein the processor system operations further comprise using the traceable path to generate accuracy confirmation information for the matching UDRSV.

8. A computer-implemented method comprising:
accessing, using a processor system, query information associated with a to-be-augmented information set (TBAIS) having a TBAIS format;
applying the query information, the TBAIS, and the TBAIS format to a sequencer to generate a sequence;
applying the sequence to a non-parametric transformer of the processor system to generate query information sequence vectors (QISV) that represent the query information and the TBAIS;
accessing, using the processor system, unannotated data repository information sequence vectors (UDRSV) that represent unannotated data repository information having a plurality of information formats;
applying an approximate nearest neighbor (ANN) search technique of the processor system to identify matching UDRSV comprising the UDRSV that match the QISV;
wherein the non-parametric transformer and the ANN search technique provide a traceable path to how the ANN search technique identified the matching UDRSV; and
generating, using the processor system, a response to the query information based at least in part on the matching UDRSV.

9. The computer-implemented method of claim 8 further comprising incorporating the response into the TBAIS to generate an augmented version of the TBAIS.

10. The computer-implemented method of claim 8 , wherein the TBAIS format is different from at least one of the plurality of information formats.

11. The computer-implemented method of claim 10 , wherein the TBAIS format is selected from the group consisting of a database, a table, and a portable document format (PDF).

12. The computer-implemented method of claim 10 , wherein the information format is selected from the group consisting of a database, a table, and a portable document format (PDF).

13. The computer-implemented method of claim 8 further comprising using the traceable path to generate an explanation of how the ANN search technique identified the matching UDRSV.

14. The computer-implemented method of claim 8 further comprising using the traceable path to generate accuracy confirmation information for the matching UDRSV.

15. A computer program product comprising a computer readable program stored on a computer readable storage medium, wherein the computer readable program, when executed on a processor system, causes the processor to perform processor system operations comprising:
accessing query information associated with a to-be-augmented information set (TBAIS) having a TBAIS format;
applying the query information, the TBAIS, and the TBAIS format to a sequencer to generate a sequence;
applying the sequence to a non-parametric transformer of the processor system to generate query information sequence vectors (QISV) that represent the query information and the TBAIS;
accessing unannotated data repository information sequence vectors (UDRSV) that represent unannotated data repository information having a plurality of information formats;
applying an approximate nearest neighbor (ANN) search technique of the processor system to identify matching UDRSV comprising the UDRSV that match the QISV;
wherein the non-parametric transformer and the ANN search technique provide a traceable path to how the ANN search technique identified the matching UDRSV; and
generating a response to the query information based at least in part on the matching UDRSV.

16. The computer program product of claim 15 , wherein the processor system operations further comprise incorporating the response into the TBAIS to generate an augmented version of the TBAIS.

17. The computer program product of claim 15 , wherein the TBAIS format is different from at least one of the plurality of information formats.

18. The computer program product of claim 17 , wherein the TBAIS format and the information format are each selected from the group consisting of a database, a table, and a portable document format (PDF).

19. The computer program product of claim 15 , wherein the processor system operations further comprise using the traceable path to generate an explanation of how the ANN search technique identified the matching UDRSV.

20. The computer program product of claim 15 , wherein the processor system operations further comprise using the traceable path to generate accuracy confirmation information for the matching UDRSV.

Assignees

• IBM
• Univ Of Massachusetts Amherst

Inventors

• Glass Michael Robert
• Rossiello Gaetano
• Gliozzo Alfio Massimiliano
• Naik Ankita Rajaram
• Wu Xueqing

Classifications

• G06F16/24578

  using ranking · CPC title

• G06F16/258

  Data format conversion from or to a database · CPC title

• G06F16/2282 Primary

  Tablespace storage structures; Management thereof · CPC title