Templated rule-based data augmentation for intent extraction

Patent metadata

Field	Value
Publication number	US-2021224485-A1
Application number	US-202117301092-A
Country	US
Kind code	A1
Filing date	Mar 24, 2021
Priority date	Mar 23, 2018
Publication date	Jul 22, 2021
Grant date	—

Abstract

An agent automation system includes a memory configured to store a natural language understanding (NLU) framework and a model, wherein the model includes at least one original meaning representation. The system includes a processor configured to execute instructions of the NLU framework to cause the agent automation system to perform actions including: performing rule-based generalization of the model to generate at least one generalized meaning representation of the model from the at least one original meaning representation of the model; performing rule-based refinement of the model to prune or modify the at least one generalized meaning representation of the model, or the at least one original meaning representation of the model, or a combination thereof; and after performing the rule-based generalization and the rule-based refinement of the model, using the model to extract intents/entities from a received user utterance

First claim

What is claimed is:

1 . An agent automation system, comprising:
at least one memory configured to store a set of generalizing rules and a model having one or more meaning representations, wherein each of the one or more meaning representations represents an utterance and includes one or more intent subtrees; and at least one processor configured to execute instructions to cause the agent automation system to perform a generalization process for each intent subtree of each meaning representation of the one or more meaning representations, the generalization process comprising:
for each active generalizing rule of the set of generalizing rules:
in response to determining that the active generalizing rule is applicable to the intent subtree of the meaning representation of the model:
applying the active generalizing rule to generate a generalized meaning representation from the meaning representation; and
updating the model to include the generalized meaning representation.

2 . The system of claim 1 , wherein the model is an understanding model and the one or more meaning representations comprise a plurality of meaning representations, each representing a sample utterance of an intent/entity model.

3 . The system of claim 1 , wherein the model is an utterance meaning model and the one or more meaning representations each represent portion of a received user utterance.

4 . The system of claim 1 , wherein the generalization process comprises:
extracting the one or more intent subtrees of the one or more meaning representations of the model.

5 . The system of claim 1 , wherein the set of generalizing rules comprise a subject/object rule configured to interchange a subject and an object of the meaning representation to generate the generalized meaning representation.

6 . The system of claim 1 , wherein the set of generalizing rules comprise a passive/active rule configured to convert the meaning representation from passive to active, or from active to passive, to generate the generalized meaning representation.

7 . The system of claim 1 , wherein the at least one memory is configured to store a set of refining rules and the at least one processor is configured to execute the instructions to cause the agent automation system to perform a refining process for each intent subtree of each meaning representation of the one or more meaning representations and each generalized meaning representation of the model, the refining process comprising:
for each active refining rule of the set of refining rules:
in response to determining that the active refining rule is applicable to the intent subtree of the meaning representation of the model:
applying the active refining rule to generate a refined meaning representation from the meaning representation, to determine that the meaning representation should be removed, or a combination thereof; and
updating the model to include the refined meaning representation, to remove the meaning representation, or a combination thereof.

8 . The system of claim 7 , wherein the set of refining rules comprise a substitution rule configured to replace a portion of the meaning representation when generating the refined meaning representation.

9 . The system of claim 8 , wherein the portion comprises one or nodes, one or more subtrees, one or more word vectors, or one or more subtree vectors, or a combination thereof.

10 . The system of claim 7 , wherein the set of refining rules comprise a pruning rule configured to determine that the meaning representation should be removed when the meaning representation is within a threshold similarity of another meaning representation of the model.

11 . The system of claim 7 , wherein the at least one processor is configured to execute the instructions to cause the agent automation system to, after performing the generalizing process and the refining process, perform actions comprising:
generating at least one meaning representation for a received user utterance; and searching the at least one meaning representation of the received user utterance within one or more meaning representations, generalized meaning representations, and/or refined meaning representations of the model to extract intents/entities from the received user utterance.

12 . A method of operating an agent automation system, comprising:
performing a generalizing process for each intent subtree of each meaning representation of one or more meaning representations of a model, the generalizing process comprising:
for each active generalizing rule of a set of generalizing rules:
in response to determining that the active generalizing rule is applicable to the intent subtree of the meaning representation of the model:
applying the active generalizing rule to generate a generalized meaning representation from the meaning representation; and
updating the model to include the generalized meaning representation.

13 . The method of claim 12 , wherein the generalization process comprises:
after generating the generalized meaning representation, in response to determining that the active generalizing rule indicates that a user should review the generalized meaning representation, queuing the generalized meaning representation for user review and awaiting user input before updating the model to include the generalized meaning representation.

14 . The method of claim 12 , comprising:
performing a refining process for each intent subtree of each meaning representation of the one or more meaning representations and each generalized meaning representation of the model, the refining process comprising: for each active refining rule of the set of refining rules:
in response to determining that the active refining rule is applicable to the intent subtree of the meaning representation of the model:
applying the active refining rule to generate a refined meaning representation from the meaning representation, to determine that the meaning representation should be removed, or a combination thereof; and
updating the model to include the refined meaning representation, to remove the meaning representation, or a combination thereof.

15 . The method of claim 14 , wherein the refining process comprises:
after generating the refined meaning representation, in response to determining that the active refining rule indicates that a user should review the refined meaning representation, queuing the refined meaning representation for user review and awaiting user input before updating the model to include the refined meaning representation.

16 . A non-transitory, computer-readable medium storing instructions that are executable by a processor of an agent automation system, the instructions comprising instructions to:
perform a generalizing process for each intent subtree of each meaning representation of one or more meaning representations of a model, the generalizing process comprising:
for each active generalizing rule of a set of generalizing rules:
in response to determining that the active generalizing rule is applicable to the intent subtree of the meaning representation of the model:
apply the active generalizing rule to generate a generalized meaning representation from the meaning representation; and
update the model to include the generalized meaning representation.

17 . The medium of claim 16 , comprising instructions to determine whether the active generalizing rule is applicable comprise instructions by:
determining whether the active generalizing rule is applicable to the model, to a particular

Assignees

• Servicenow Inc

Inventors

• Sapugay Edwin
• Madamala Anil Kumar
• Naboka Maxim
• Sunkara Srinivas Satyasai
• Santos Lewis Savio Landry
• Subbarao Murali B

Classifications

• G06F40/30 Primary

  Semantic analysis · CPC title

• G10L25/48

  specially adapted for particular use · CPC title

• G10L15/22

  Procedures used during a speech recognition process, e.g. man-machine dialogue · CPC title

• G06F40/205

  Parsing · CPC title

• G10L15/1822

  Parsing for meaning understanding · CPC title