SUMMARY GENERATING DEVICE, SUMMARY GENERATING METHOD, AND INFORMATION STORAGE MEDIUM

• US 20200134011A1
• Filed: 10/16/2019
• Published: 04/30/2020
• Est. Priority Date: 10/26/2018
• Status: Active Grant

• Current Assignee: Rakuten Group, Inc.
• Original Assignee: Rakuten, Inc. (Rakuten Group, Inc.)
• Inventors: DAULTANI, Vijay, NIO, Lasguido, CHUNG, Youngjoo
• Granted Patent: US 11,061,950 B2
• Time in Patent Office: Days
• CPC Class Codes: G06F 16/3334 Selection or weighting of terms from queries, including natural language queries
• CPC Class Codes: G06F 16/345 Summarisation for human users
• CPC Class Codes: G06F 40/216 using statistical methods
• CPC Class Codes: G06F 40/253 Grammatical analysis; Style critique
• CPC Class Codes: G06F 40/30 Semantic analysis

Abstract

A summary generating device includes at least one processor that is configured to use a plurality of different algorithms, which extract one or more elements from a document and obtain an appearance degree of each of the extracted elements, so as to obtain the elements and the respective appearance degrees of the elements from the document, normalize the obtained appearance degrees for each of the algorithms, select at least one sentence from the document based on the normalized appearance degrees, and generate a summary of the document based on the selected sentence.

Claims

1. A summary generating device comprising at least one processor is configured to:

use a plurality of different algorithms, which extract one or more elements from a document and obtain an appearance degree of each of the extracted elements, so as to obtain the elements and the respective appearance degrees of the elements from the document;

normalize the obtained appearance degrees for each of the algorithms;

select at least one sentence from the document based on the normalized appearance degrees; and

generate a summary of the document based on the selected sentence.

2. The summary generating device according to claim 1, wherein

the at least one processor:

filters the elements based on the normalized appearance degrees, and

uses the normalized appearance degrees of the filtered elements so as to select at least one sentence from the document based on a predetermined select condition.

3. The summary generating device according to claim 2, wherein

the predetermined select condition is where a sum total of the normalized appearance degrees of the elements included in the selected sentence is maximized and an amount of the selected sentence is less than a predetermined amount.

4. The summary generating device according to claim 3, wherein

the at least one processor uses an integer linear program so as to select at least one sentence from the document.

5. The summary generating device according to claim 2, wherein

the at least one processor determines a filtering condition based on a distribution of the normalized appearance degrees, and performs filtering based on the determined filtering condition.

6. The summary generating device according to claim 2, wherein

the at least one processor performs the filtering such that a predetermined percentage of the elements having higher normalized appearance degrees are extracted, regardless of an algorithm used for extraction.

7. The summary generating device according to claim 1, wherein

the plurality of algorithms include at least one of a TF-IDF (Term frequency-Inverse document frequency) method or an attribute extraction method for extracting an attribute value of a predetermined attribute as the element.

8. The summary generating device according to claim 1, wherein

the at least one processor generates the summary by arranging the selected sentences in an order that the sentences appear in the document.

9. The summary generating device according to claim 1, wherein

the document is a description of a product, a facility, or service,

the plurality of algorithms include the attribute extraction method for extracting an attribute value of an attribute of the product, the facility, or the service as the element, and

the at least one processor performs normalization such that a distribution of the appearance degrees obtained by the attribute extraction method is close to a distribution of the normalized appearance degrees obtained by other algorithms.

10. A summary generating method comprising:

using a plurality of different algorithms, which extract one or more elements from a document and obtain an appearance degree of each of the extracted elements, so as to obtain the elements and the respective appearance degrees of the elements from the document;

normalizing the obtained appearance degrees for each of the algorithms;

selecting at least one sentence from the document based on the normalized appearance degrees; and

generating a summary of the document based on the selected sentence.

11. A non-transitory computer-readable information storage medium for storing a program that causes a computer to:

use a plurality of different algorithms, which extract one or more elements from a document and obtain an appearance degree of each of the extracted elements, so as to obtain the elements and the respective appearance degrees of the elements from the document;

normalize the obtained appearance degrees for each of the algorithms;

select at least one sentence from the document based on the normalized appearance degrees; and

generate a summary of the document based on the selected sentence.