QUERY-BASED EXTRACTIVE SUMMARIZATION

• US 20200134091A1
• Filed: 10/29/2018
• Published: 04/30/2020
• Est. Priority Date: 10/29/2018
• Status: Active Grant

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Feigenblat, GUY, Konopnicki, David, Roitman, Haggai
• Granted Patent: US 10,831,806 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 16/338 Presentation of query results
• CPC Class Codes: G06F 16/345 Summarisation for human users
• CPC Class Codes: G06F 16/93 Document management systems

Abstract

A system comprising at least one hardware processor; and a non-transitory computer-readable storage medium having stored thereon program instructions executable to receive, as input, one or more digital documents, a query statement, and a summary length constraint, automatically generate, for each of said one or more digital documents, an initial summary based, at least in part, on a first sentence selection which satisfies said query statement, a modified said summary length constraint, and a first summary quality goal, automatically extract, from each of said initial summaries, one or more associated feedback metrics, and automatically generate, for each of said one or more digital documents, a final summary based, at least in part, on: (i) a second sentence selection which satisfies said query statement, said summary length constraint, and a second summary quality goal, and (ii) at least one of said associated feedback metrics.

Claims

1. A system comprising:

at least one hardware processor; and

a non-transitory computer-readable storage medium having stored thereon program instructions, the program instructions executable by the at least one hardware processor to;

receive, as input, one or more digital documents, a query statement, and a summary length constraint,

automatically generate, for each of said one or more digital documents, an initial summary based, at least in part, on a first sentence selection which satisfies said query statement, a modified said summary length constraint, and a first summary quality goal,

automatically extract, from each of said initial summaries, one or more associated feedback metrics, and

automatically generate, for each of said one or more digital documents, a final summary based, at least in part, on;

(i) a second sentence selection which satisfies said query statement, said summary length constraint, and a second summary quality goal, and

(ii) at least one of said associated feedback metrics.

2. The system of claim 1, wherein said one or more associated feedback metrics are selected from the group consisting of:

frequency of unique textual aspects, distribution of unique textual aspects, patterns of unique textual aspects, relative start-positions of sentences, maximum sentence length, minimum sentence length, average sentence length, maximum number of words in each sentence, minimum number of words in each sentence, and average number of words in each sentence.

3. The system of claim 1, wherein said unique textual aspects comprise at least one of n-grams, entities, and concepts.

4. The system of claim 1, wherein said summary length constraint is expressed as number of words.

5. The system of claim 1, wherein said modified summary length constraint is longer than said summary length constraint.

6. The system of claim 1, wherein said first quality goals is summary saliency, and said second summary quality goal is summary focus.

7. The system of claim 6, wherein said first quality goal comprises one or more quality goals selected from the group consisting of:

document coverage, sentence position bias, summary length, asymmetric coverage, and focus drift.

8. The system of claim 6, wherein said second quality goal comprises a query-relevancy quality goal.

9. The system of claim 1, wherein said first and second sentence selections are each based, at least in part, on solving an optimization problem using one or more specified optimization frameworks.

10. The system of claim 9, wherein said optimization framework is a cross-entropy framework.

11. The system of claim 1, wherein said modified summary length constraint is an adaptive summary length constraint, based, at least in part, on importance sampling.

12. A method comprising:

operating at least one hardware processor for;

receiving, as input, one or more digital documents, a query statement, and a summary length constraint,

automatically generating, for each of said one or more digital documents, an initial summary based, at least in part, on a first sentence selection which satisfies said query statement, a modified said summary length constraint, and a first summary quality goal,

automatically extracting, from each of said initial summaries, one or more associated feedback metrics, and

automatically generating, for each of said one or more digital documents, a final summary based, at least in part, on;

(i) a second sentence selection which satisfies said query statement, said summary length constraint, and a second summary quality goal, and

(ii) at least one of said associated feedback metrics.

13. The method of claim 12, wherein said one or more associated feedback metrics are selected from the group consisting of:

frequency of unique textual aspects, distribution of unique textual aspects, patterns of unique textual aspects, relative start-positions of sentences, maximum sentence length, minimum sentence length, average sentence length, maximum number of words in each sentence, minimum number of words in each sentence, and average number of words in each sentence.

14. The method of claim 12, wherein said unique textual aspects comprise at least one of n-grams, entities, and concepts.

15. The method of claim 12, wherein said summary length constraint is expressed as number of words.

16. The method of claim 12, wherein said modified summary length constraint is longer than said summary length constraint.

17. The method of claim 12, wherein said first quality goals is summary saliency, and said second summary quality goal is summary focus.

18. The method of claim 17, wherein said first quality goal comprises one or more quality goals selected from the group consisting of:

document coverage, sentence position bias, summary length, asymmetric coverage, and focus drift.

19. The method of claim 17, wherein said second quality goal comprises a query-relevancy quality goal.

20. The method of claim 12, wherein said first and second sentence selections are each based, at least in part, on solving an optimization problem using one or more specified optimization frameworks.