Information processing apparatus

Patent metadata

Field	Value
Publication number	US-11699038-B2
Application number	US-202117177819-A
Country	US
Kind code	B2
Filing date	Feb 17, 2021
Priority date	Sep 18, 2020
Publication date	Jul 11, 2023
Grant date	Jul 11, 2023

Abstract

An information processing apparatus includes a processor configured to input an arbitrary text and plural answer candidates extracted from the text to a question generator that generates a question sentence associated with an answer, cause the question generator to generate a question sentence associated with the text and the plural answer candidates, and cause the question generator to output generated data including a set of the plural answer candidates and the question sentence.

First claim

What is claimed is:

1. An information processing apparatus comprising
a processor configured to
input an arbitrary text and a plurality of answer candidates extracted from the text to a question generator,
cause the question generator to generate a question sentence associated with the text and the plurality of answer candidates, and
cause the question generator to output generated data including a set of the plurality of answer candidates and the question sentence, wherein

the processor is further configured to
train a checking model, that checks the generated data, by using appropriate question/answer data extracted from a single-answer data set, and inappropriate question/answer data created from the single-answer data set,
determine whether the generated data is appropriate by using the checking model, and
output the generated data only if the generated data is appropriate.

2. The information processing apparatus according to claim 1 , wherein the processor is configured to use the generated data as supervised data for a multi-answer reading comprehension model that extracts a plurality of answers from a question and analysis target data, and outputs the plurality of extracted answers.

3. The information processing apparatus according to claim 2 , wherein the processor is configured to train the multi-answer reading comprehension model by using the generated data as the supervised data.

4. The information processing apparatus according to claim 3 , wherein the processor is configured to train the multi-answer reading comprehension model by further using the single-answer data set as the supervised data.

5. The information processing apparatus according to claim 1 , wherein the processor is configured to determine whether each of the plurality of answer candidates in the generated data is appropriate by using the checking model.

6. The information processing apparatus according to claim 1 , wherein the processor is configured to train the question generator by using the single-answer data set.

7. An information processing apparatus comprising:
a processor configured to
input an arbitrary text and at least one answer candidate extracted from the text to a question generator,
cause the question generator to generate a question sentence associated with the text and the answer candidate,
cause the question generator to output generated data including a set of the text and the question sentence,
input the generated data to a machine reading comprehension model that extracts a plurality of answers to the question sentence from the text, and outputs the extracted answers,
cause the machine reading comprehension model to generate the plurality of answers to the question sentence in the input generated data, and
cause the machine reading comprehension model to output the generated data with the plurality of answers added to the generated data, wherein

the processor is further configured to
if the plurality of answers output from the machine reading comprehension model include different answers, use the generated data output from the machine reading comprehension model as supervised data for a multi-answer reading comprehension model that extracts a plurality of answers from a question and analysis target data, and outputs the plurality of extracted answers,
train a checking model, that checks the generated data, by using appropriate question/answer data extracted from a single-answer data set, and inappropriate question/answer data created from the single-answer data set,
determine whether the generated data is appropriate by using the checking model, and
use the generated data as the supervised data for the multi-answer reading comprehension model only if the generated data is appropriate.

8. The information processing apparatus according to claim 7 , wherein the processor is configured to train the multi-answer reading comprehension model by using the generated data output from the machine reading comprehension model as the supervised data.

9. The information processing apparatus according to claim 8 , wherein the processor is configured to train the multi-answer reading comprehension model by further using the single-answer data set as the supervised data.

10. The information processing apparatus according to claim 9 , wherein the processor is configured to
replace the trained multi-answer reading comprehension model with the machine reading comprehension model, and
input data output from the question generator to the machine reading comprehension model.

11. The information processing apparatus according to claim 7 , wherein the processor is configured to determine whether each of a plurality of the answer candidates in the generated data is appropriate by using the checking model.

12. The information processing apparatus according to claim 7 , wherein the machine reading comprehension model is configured to extract the answers from the question sentence in the input generated data by using a plurality of the single-answer models prepared in advance.

13. The information processing apparatus according to claim 7 , wherein the processor is configured to train the question generator by using the single-answer data set.

14. An information processing apparatus comprising
a processor configured to
input an arbitrary text and at least one answer candidate extracted from the text to a question generator,
cause the question generator to generate a question sentence associated with the text and the answer candidate,
cause the question generator to output generated data including a set of the text and the question sentence,
input the generated data to a machine reading comprehension model that extracts a plurality of answers to the question sentence from the text, and outputs the extracted answers,
cause the machine reading comprehension model to generate the plurality of answers to the question sentence in the input generated data, and
cause the machine reading comprehension model to output the generated data with the plurality of answers added to the generated data, wherein

the processor is further configured to
if the plurality of answers output from the machine reading comprehension model include different answers, use the generated data output from the machine reading comprehension model as supervised data for a multi-answer reading comprehension model that extracts a plurality of answers from a question and analysis target data, and outputs the plurality of extracted answers,
train the multi-answer reading comprehension model by using the generated data output from the machine reading comprehension model and a single-answer data set prepared in advance as the supervised data,
replace the trained multi-answer reading comprehension model with the machine reading comprehension model, and
input data output from the question generator to the machine reading comprehension model.

Assignees

• Fujifilm Business Innovation Corp

Inventors

• Takahashi Takumi
• Taniguchi Motoki
• Taniguchi Tomoki
• Ohkuma Tomoko

Classifications

• G06F40/279

  Recognition of textual entities · CPC title

• G06N20/00

  Machine learning · CPC title

• G06F40/205

  Parsing · CPC title

• G06F40/289

  Phrasal analysis, e.g. finite state techniques or chunking · CPC title

• G06F40/284 Primary

  Lexical analysis, e.g. tokenisation or collocates · CPC title