Reduced training for dialog systems using a database

Patent metadata

Field	Value
Publication number	US-2021082425-A1
Application number	US-202016983950-A
Country	US
Kind code	A1
Filing date	Aug 3, 2020
Priority date	Sep 12, 2019
Publication date	Mar 18, 2021
Grant date	—

Abstract

Techniques are described for training and executing a machine learning model using data derived from a database. A dialog system uses data from the database to generate related training data for natural language understanding applications. The generated training data is then used to train a machine learning model. This enables the dialog system to leverage a large amount of available data to speed up the training process as compared to conventional labeling techniques. The dialog system uses the trained machine learning model to identify a named entity from a received spoken utterance and generate and output a speech response based upon the identified named entity.

First claim

What is claimed is:

1 . A method comprising:
receiving, by a dialog system, a spoken utterance; identifying, by the dialog system, a named entity from the spoken utterance using a machine learning model, wherein the machine learning model has been trained on data extracted from a database that maps a plurality of named entities to respective named entity types based upon columns in the database; generating, by the dialog system, a speech response based upon the identified named entity; and providing, by the dialog system, the speech response as output.

2 . The method of claim 1 , further comprising:
extracting raw data from the database; generating training data from the extracted raw data; and training the machine learning model on the generated training data.

3 . The method of claim 2 , wherein generating the training data comprises:
identifying metadata associated with the columns of the database; and using the metadata and corresponding entries of the columns as seed data to generate the training data.

4 . The method of claim 1 , wherein:
the machine learning model is a first machine learning model and the named entity is a first named entity; and the method further comprises identifying a second named entity using a second machine learning model.

5 . The method of claim 1 , wherein:
the database further comprises a plurality of requestable values; and the method further comprises: identifying, by the dialog system using the database, a requestable value, of the plurality of requestable values, that maps to the identified named entity, wherein the speech response includes the requestable value or a derivative thereof.

6 . The method of claim 5 , wherein:
the database comprises a plurality of tables; and the method further comprises selecting a particular table from the plurality of tables based upon the identified named entity,
wherein the selected table is used to identify the requestable value.

7 . The method of claim 6 , wherein identifying the requestable value comprises executing a query on the selected table to retrieve the requestable value mapped to the identified named entity.

8 . A non-transitory computer-readable memory storing a plurality of instructions executable by one or more processors, the plurality of instructions comprising instructions that when executed by the one or more processors cause the one or more processors to perform processing comprising:
receiving a spoken utterance; identifying a named entity from the spoken utterance using a machine learning model, wherein the machine learning model has been trained on data extracted from a database that maps a plurality of named entities to respective named entity types based upon columns in the database; generating a speech response based upon the identified named entity; and providing the speech response as output.

9 . The non-transitory computer-readable memory of claim 8 , the processing further comprising:
extracting raw data from the database; generating training data from the extracted raw data; and training the machine learning model on the generated training data.

10 . The non-transitory computer-readable memory of claim 9 , wherein generating the training data comprises:
identifying metadata associated with the columns of the database; and using the metadata and corresponding entries of the columns as seed data to generate the training data.

11 . The non-transitory computer-readable memory of claim 8 , wherein:
the machine learning model is a first machine learning model and the named entity is a first named entity; and the processing further comprises identifying a second named entity using a second machine learning model.

12 . The non-transitory computer-readable memory of claim 8 , wherein:
the database further comprises a plurality of requestable values; and the processing further comprises: identifying, using the database, a requestable value, of the plurality of requestable values, that maps to the identified named entity, wherein the speech response includes the requestable value or a derivative thereof.

13 . The non-transitory computer-readable memory of claim 12 , wherein:
the database comprises a plurality of tables; and the processing further comprises selecting a particular table from the plurality of tables based upon the identified named entity,
wherein the selected table is used to identify the requestable value.

14 . The non-transitory computer-readable memory of claim 13 , wherein identifying the requestable value comprises executing a query on the selected table to retrieve the requestable value mapped to the identified named entity.

15 . A dialog system comprising:
one or more processors; a memory coupled to the one or more processors, the memory storing a plurality of instructions executable by the one or more processors, the plurality of instructions comprising instructions that when executed by the one or more processors cause the one or more processors to perform processing comprising: receiving a spoken utterance; identifying a named entity from the spoken utterance using a machine learning model, wherein the machine learning model has been trained on data extracted from a database that maps a plurality of named entities to respective named entity types based upon columns in the database; generating a speech response based upon the identified named entity; and providing the speech response as output.

16 . The dialog system of claim 15 , the processing further comprising:
extracting raw data from the database; generating training data from the extracted raw data; and training the machine learning model on the generated training data.

17 . The dialog system of claim 16 , wherein generating the training data comprises:
identifying metadata associated with the columns of the database; and using the metadata and corresponding entries of the columns as seed data to generate the training data.

18 . The dialog system of claim 15 , wherein:
the machine learning model is a first machine learning model and the named entity is a first named entity; and the processing further comprises identifying a second named entity using a second machine learning model.

19 . The dialog system of claim 15 , wherein:
the database further comprises a plurality of requestable values; and the processing further comprises: identifying, using the database, a requestable value, of the plurality of requestable values, that maps to the identified named entity, wherein the speech response includes the requestable value or a derivative thereof.

20 . The dialog system of claim 19 , wherein:
the database comprises a plurality of tables; and the processing further comprises selecting a particular table from the plurality of tables based upon the identified named entity,
wherein the selected table is used to identify the requestable value.

Assignees

• Oracle Int Corp

Inventors

• Johnson Mark Edward
• Kennewick Michael Rye

Classifications

• G06N3/0895

  Weakly supervised learning, e.g. semi-supervised or self-supervised learning · CPC title

• G06N3/09

  Supervised learning · CPC title

• G06N3/096

  Transfer learning · CPC title

• G06N3/08

  Learning methods · CPC title

• G06N20/00

  Machine learning · CPC title