Machine learning for training NLP agent

Patent metadata

Field	Value
Publication number	US-12014142-B2
Application number	US-202117354825-A
Country	US
Kind code	B2
Filing date	Jun 22, 2021
Priority date	Jun 22, 2021
Publication date	Jun 18, 2024
Grant date	Jun 18, 2024

Abstract

A computer-implemented process for training a natural language processing (NLP) agent having a reinforced learning model includes the following operations. A type of document from a document corpus is identified using metadata particularly associated with the document. The NLP agent tokenizes the document to generate a plurality of tokens. Using a schema identified from the type of the document, one of the plurality of tokens is compared to a system of record (SOR) field from the schema. A similarity score between the one of the plurality of tokens with a correct value and a reward based upon the similarity score are generated. A determination is made that an optimum minimum average similarity rate has not been obtained. Based upon the determination, the reinforced learning model is trained using a loss function that includes the reward.

First claim

What is claimed is:

1. A computer-implemented method for training a natural language processing (NLP) agent having a reinforced learning model, comprising:
identifying, using metadata particularly associated with a document from a document corpus, a type of the document;
tokenizing, by the NLP agent, the document to generate a plurality of tokens;
comparing one of the plurality of tokens to a system of record (SOR) field from a schema identified from the type of the document;
generating a similarity score between the one of the plurality of tokens with a correct value and a reward based upon the similarity score;
determining that an optimum minimum average similarity rate has not been obtained; and
training, based upon the determining, the reinforced learning model using a loss function that includes the reward.

2. The method of claim 1 , wherein
the comparing and the generating are performed for all of a plurality of SOR fields in the schema.

3. The method of claim 2 , wherein
the determining is based upon similarity scores for all of the plurality of SOR fields.

4. The method of claim 1 , wherein
the reinforced learning model is a Deep Q Network.

5. The method of claim 1 , wherein
the similarity score use a Levenshtein distance.

6. The method of claim 1 , wherein
after a determination is made that optimum minimum average similarity rate has been obtained, the training is repeated using another document from the document corpus.

7. The method of claim 1 , wherein
the tokenizing includes identifying a respective location for each of the plurality of tokens.

8. The method of claim 1 , wherein
the NLP agent includes a machine learning engine.

9. A computer hardware system for training a natural language processing (NLP) agent having a reinforced learning model, comprising:
a hardware processor configured to perform the following executable operations:
identifying, using metadata particularly associated with a document from a document corpus, a type of the document;
tokenizing, by the NLP agent, the document to generate a plurality of tokens;
comparing one of the plurality of tokens to a system of record (SOR) field from a schema identified from the type of the document;
generating a similarity score between the one of the plurality of tokens with a correct value and a reward based upon the similarity score;
determining that an optimum minimum average similarity rate has not been obtained; and
training, based upon the determining, the reinforced learning model using a loss function that includes the reward.

10. The system of claim 9 , wherein
the comparing and the generating are performed for all of a plurality of SOR fields in the schema.

11. The system of claim 10 , wherein
the determining is based upon similarity scores for all of the plurality of SOR fields.

12. The system of claim 9 , wherein
the reinforced learning model is a Deep Q Network.

13. The system of claim 9 , wherein
the similarity score use a Levenshtein distance.

14. The system of claim 9 , wherein
after a determination is made that optimum minimum average similarity rate has been obtained, the training is repeated using another document from the document corpus.

15. The system of claim 9 , wherein
the tokenizing includes identifying a respective location for each of the plurality of tokens.

16. The system of claim 9 , wherein
the NLP agent includes a machine learning engine.

17. A computer program product, comprising:
a computer readable storage medium having stored therein program code for training a natural language processing (NLP) agent having a reinforced learning model,
the program code, which when executed by a computer hardware system, cause the computer hardware system to perform:
identifying, using metadata particularly associated with a document from a document corpus, a type of the document;
tokenizing, by the NLP agent, the document to generate a plurality of tokens;
comparing one of the plurality of tokens to a system of record (SOR) field from a schema identified from the type of the document;
generating a similarity score between the one of the plurality of tokens with a correct value and a reward based upon the similarity score;
determining that an optimum minimum average similarity rate has not been obtained; and
training, based upon the determining, the reinforced learning model using a loss function that includes the reward.

18. The computer program product of claim 17 , wherein
the comparing and the generating are performed for all of a plurality of SOR fields in the schema; and
the determining is based upon similarity scores for all of the plurality of SOR fields.

19. The computer program product of claim 17 , wherein
the reinforced learning model is a Deep Q Network, and
the NLP agent includes a machine learning engine.

20. The computer program product of claim 17 , wherein
after a determination is made that optimum minimum average similarity rate has been obtained, the training is repeated using another document from the document corpus.

Assignees

• IBM

Inventors

• Diamanti Gary Francis
• Kwatra Shikhar
• Anderson Ryan
• Goulart Silva Rodrigo

Classifications

• G06N3/0499

  Feedforward networks · CPC title

• G06N3/092

  Reinforcement learning · CPC title

• G06F18/22 Primary

  Matching criteria, e.g. proximity measures · CPC title

• G06N20/00

  Machine learning · CPC title

• G06N3/08

  Learning methods · CPC title