Granted Patent B2

US 12,373,083 · App. 17/169,825 · Granted Jul 29, 2025

Inventors: Dan G. Tecuci (Austin, TX); Ravi Kiran Reddy Palla (Sunnyvale, CA); Hamid Reza Motahari Nezhad (Los Altos, CA); Vincent Poon (Millbrae, CA); Nigel Paul Duffy (San Francisco, CA); Joseph Nipko (Austin, TX)

G06F3/0482G06F18/2148G06F18/2178G06N3/02G06N20/20G06V10/22G06V10/771G06V10/82G06V10/987G06V30/19167G06V30/19173G06V30/40G06V30/43

Abstract

Claims (53)

1. A computer-implemented method, comprising:

generating a dataset including a plurality of value pairs, each value pair from the plurality of value pairs including an error-free value and an error-containing value, at least one error-free value included in at least one value pair from the plurality of value pairs modified to generate at least one error-containing value included in the at least one value pair from the plurality of value pairs;

training a machine learning model with the dataset to produce a trained machine learning model, training the machine learning model including

applying a semantic model included in the machine learning model and not a similarity model included in the machine learning model,

applying, after applying the semantic model and not the similarity model, the similarity model and the semantic model, and

in response to applying the semantic model and the similarity model, applying the similarity model and a multi-layer perceptron model included in the machine learning model without applying the semantic model;

detecting, via the trained machine learning model, an error in an electronically-stored file;

in response to detecting the error, converting the electronically-stored file, via the trained machine learning model, into a modified electronically-stored file that does not include the error;

identifying a set of objects associated with the modified electronically-stored file; and

in response to a user interaction with a representation of an object from the set of objects, updating the trained machine learning model based on the user interaction, to have more weight for at least one of retrieval or ranking, and to have less weight for semantic searching.

2. The computer-implemented method of claim 1 , wherein the error is an optical character recognition (OCR) induced error.

3. The computer-implemented method of claim 1 , wherein the machine learning model includes at least one of a sequence-to-sequence model or an attention algorithm.

4. The computer-implemented method of claim 1 , wherein the converting the electronically-stored file into the modified electronically-stored file is based on a value pair (1) from the plurality of value pairs and (2) associated with the detected error.

5. The computer-implemented method of claim 4 , wherein the converting the electronically-stored file into the modified electronically-stored file includes replacing an erroneous data segment of the electronically-stored file with the error-free value of the value pair from the plurality of value pairs, the erroneous data segment associated with the detected error.

6. The computer-implemented method of claim 1 , wherein the machine learning model operates within an artificial neural network.

7. The computer-implemented method of claim 1 , wherein the detecting the error in the electronically-stored file includes:

receiving, from a named-entity recognition (NER) model, a signal representing a non-extracted data string that is associated with the electronically-stored file; and

inspecting the non-extracted data string to identify the error.

8. The computer-implemented method of claim 1 , wherein each of the electronically-stored file and the modified electronically-stored file represents an electronic document.

9. An apparatus, comprising:

a processor; and

a memory storing instructions, executable by the processor, to:

generate a dataset including a plurality of value pairs, each value pair from the plurality of value pairs including an error-free value and an error-containing value, at least one error-free value included in at least one value pair from the plurality of value pairs modified to generate at least one error-containing value included in the at least one value pair from the plurality of value pairs;

train a machine learning model with the dataset to produce a trained machine learning model, training the machine learning model including

applying a semantic model included in the machine learning model and not a similarity model included in the machine learning model,

applying, after applying the semantic model and not the similarity model, the similarity model and the semantic model, and

in response to applying the semantic model and the similarity model, applying the similarity model and a multi-layer perceptron model included in the machine learning model without applying the semantic model;

detect, via the trained machine learning model, an error in an electronically-stored file;

in response to detecting the error, convert the electronically-stored file into a modified electronically-stored file that does not include the error;

identify a set of objects associated with the modified electronically-stored file; and

in response to a user interaction with a representation of an object from the set of objects, update the trained machine learning model based on the user interaction, to have more weight for at least one of retrieval or ranking, and to have less weight for semantic searching.

10. The apparatus of claim 9 , wherein the error is an optical character recognition (OCR) induced error.

11. The apparatus of claim 9 , wherein the machine learning model includes at least one of a sequence-to-sequence model or an attention algorithm.

12. The apparatus of claim 9 , wherein the instructions to convert the electronically-stored file into the modified electronically-stored file that does not include the error include instructions to convert the electronically-stored file into the modified electronically-stored file based on a value pair ( 1 ) from the plurality of value pairs and ( 2 ) associated with the detected error.

13. The apparatus of claim 9 , wherein the instructions to convert the electronically-stored file into the modified electronically-stored file that does not include the error include instructions to replace a data segment of the electronically-stored file with the error-free value of the value pair from the plurality of value pairs, the data segment associated with the detected error.

14. The apparatus of claim 9 , wherein the machine learning model operates within an artificial neural network.

15. The apparatus of claim 9 , wherein the instructions to detect the error in the electronically-stored file include instructions to:

receive, from a named-entity recognition (NER) model, a signal representing a non-extracted data string that is associated with the electronically-stored file; and

inspect the non-extracted data string to identify the error.

16. The apparatus of claim 9 , wherein each of the electronically-stored file and the modified electronically-stored file represents an electronic document.

17. A non-transitory, processor-readable medium storing instructions to cause a processor to:

generate a plurality of value pairs, each value pair from the plurality of value pairs including an error-free value and an error-containing value, at least one error-free value included in at least one value pair from the plurality of value pairs modified to generate at least one error-containing value included in the at least one value pair from the plurality of value pairs;

train a machine learning model with the plurality of value pairs to produce a trained machine learning model, training the machine learning model including

applying a semantic model included in the machine learning model and not a similarity model included in the machine learning model,

applying, after applying the semantic model and not the similarity model, the similarity model and the semantic model, and

in response to applying the semantic model and the similarity model, applying the similarity model and a multi-layer perceptron model included in the machine learning model without applying the semantic model;

detect, via the trained machine learning model, an error in an electronically-stored file;

in response to detecting the error, convert the electronically-stored file, via the trained machine learning model, into a modified electronically-stored file that does not include the error;

identify a set of objects associated with the modified electronically-stored file; and

in response to a user interaction with a representation of an object from the set of objects, update the trained machine learning model based on the user interaction, to have more weight for at least one of retrieval or ranking, and to have less weight for semantic searching.

18. The non-transitory, processor-readable medium of claim 17 , wherein the error is an optical character recognition (OCR) induced error.

19. The non-transitory, processor-readable medium of claim 17 , wherein the machine learning model includes at least one of a sequence-to-sequence model or an attention algorithm.

20. The non-transitory, processor-readable medium of claim 17 , wherein the instructions to convert the electronically-stored file into the modified electronically-stored file that does not include the error include instructions to replace a data segment of the electronically-stored file with the error-free value of the value pair from the plurality of value pairs, the data segment associated with the detected error.

Continuity (3)

References Cited (177)