OBTAINING DATA FROM ELECTRONIC DOCUMENTS

• US 20130282361A1
• Filed: 04/20/2012
• Published: 10/24/2013
• Est. Priority Date: 04/20/2012
• Status: Active Grant

• Current Assignee: SAP SE
• Original Assignee: SAP AG (SAP SE)
• Inventors: Hartl, Florian, Miao, Yingjie
• Granted Patent: US 9,348,811 B2
• Time in Patent Office: Days
• US Class Current: 704/9
• CPC Class Codes: G06F 40/295 Named entity recognition

Abstract

Techniques for obtaining information from an electronic document include accessing a set of related electronic documents; identifying a product page associated with the set of related electronic documents using a page recognition model, the product page comprising a plurality of terms; filtering the plurality of terms into a first set of terms and a second set of terms, the first set of terms and the second set of terms including different terms of the plurality of terms, each term in the first set of terms identified as potentially being associated with a product name, and each term in the second set of terms identified as not being associated with a product name; and identifying each term in the first set of terms as being associated with a product name or not being associated with a product name with a name recognition model.

Claims

1. A method performed with a computing system for obtaining information from a set of related electronic documents, the method comprising:

accessing the set of related electronic documents;

identifying a product page associated with the set of related electronic documents using a page recognition model, the page recognition model generated based on a first machine learning algorithm, and the product page comprising a plurality of terms;

filtering the plurality of terms into a first set of terms and a second set of terms, the first set of terms and the second set of terms including different terms of the plurality of terms, each term in the first set of terms identified as potentially being associated with a product name, and each term in the second set of terms identified as not being associated with a product name; and

identifying each term in the first set of terms as being associated with a product name or not being associated with a product name with a name recognition model, the name recognition model generated based on a second machine learning algorithm.

2. The method of claim 1, wherein identifying a product page associated with the set of related electronic documents using a page recognition model comprises generating the page recognition model by:

accessing an electronic document;

determining features associated with the electronic document;

receiving input specifying a label for the electronic document; and

generating the page recognition model using the labeled electronic document and the features as a training set.

3. The method of claim 1, wherein identifying each term in the first set of terms as being associated with a product name or not being associated with a product name with a name recognition model comprises generating the name recognition model by:

extracting one or more terms from an electronic document;

determining features associated with the one or more terms;

receiving input specifying a label for the one or more terms; and

generating the name recognition model using the labeled terms and the features as a training set.

4. The method of claim 1, wherein identifying a product page comprises:

accessing an electronic document associated with the set of related electronic documents;

determining features associated with the electronic document; and

analyzing the features to determine whether the electronic document is a product page using the page recognition model.

5. The method of claim 4, wherein determining features associated with the electronic document comprises:

analyzing a uniform resource locator, structure, and content associated with the electronic document.

6. The method of claim 5, wherein analyzing the structure associated with the electronic document comprises:

determining at least one of amount of text, length of the longest text section, length of link texts, average link length, average list entry length, average table entry length, number of links, images, tables, or markup language tags, headlines of tables or lists, or a link associated with an image, list, table, or headline in the electronic document.

7. The method of claim 1, wherein filtering the plurality of terms into a first set of terms and a second set of terms comprises:

parsing text in the product page into the plurality of terms;

for each of the plurality of terms, determining a type of the term; and

identifying the term as belonging in the first set of terms or the second set of terms based on the type of the term.

8. The method of claim 7, wherein the type of the term comprises at least one of a noun, a verb, an adverb, an adjective, a number, or a symbol.

9. The method of claim 1, wherein identifying each term of the first set in terms as being associated with a product name or not being associated with the product name comprises:

for each term in the first set of terms, determining one or more features associated with the term and a noun phrase that includes the term; and

determining, with the name recognition model, whether the term is associated with the product name or not associated with the product name based on the one or more features.

10. The method of claim 9, wherein determining one or more features associated with the term comprises:

parsing text in the product page into one or more segments;

for each segment, determining a type of the segment; and

associating the type of the segment with the term.

11. The method of claim 10, wherein the type of the segment comprises at least one of a link, a headline, a list entry, a table entry, or a sentence.

12. A computer storage medium encoded with a computer program, the program comprising instructions that when executed by one or more computers cause the one or more computers to perform operations comprising:

accessing a set of related electronic documents;

identifying a product page associated with the set of related electronic documents using a page recognition model, the page recognition model generated based on a first machine learning algorithm, and the product page comprising a plurality of terms;

filtering the plurality of terms into a first set of terms and a second set of terms, the first set of terms and the second set of terms including different terms of the plurality of terms, each term in the first set of terms identified as potentially being associated with a product name, and each term in the second set of terms identified as not being associated with a product name; and

identifying each term in the first set of terms as being associated with a product name or not being associated with a product name with a name recognition model, the name recognition model generated based on a second machine learning algorithm.

13. The computer storage medium of claim 12, wherein identifying a product page associated with the set of related electronic documents using a page recognition model comprises generating the page recognition model by:

accessing an electronic document;

determining features associated with the electronic document;

receiving input specifying a label for the electronic document; and

generating the page recognition model using the labeled electronic document and the features as a training set.

14. The computer storage medium of claim 12, wherein identifying each term in the first set of terms as being associated with a product name or not being associated with a product name with a name recognition model comprises generating the name recognition model by:

extracting one or more terms from an electronic document;

determining features associated with the one or more terms;

receiving input specifying a label for the one or more terms; and

generating the name recognition model using the labeled terms and the features as a training set.

15. The computer storage medium of claim 12, wherein identifying a product page comprises:

accessing an electronic document associated with the set of related electronic documents;

determining features associated with the electronic document; and

analyzing the features to determine whether the electronic document is a product page using the page recognition model.

16. The computer storage medium of claim 15, wherein determining features associated with the electronic document comprises:

analyzing a uniform resource locator, structure, and content associated with the electronic document.

17. The computer storage medium of claim 16, wherein analyzing the structure associated with the electronic document comprises:

determining at least one of amount of text, length of the longest text section, length of link texts, average link length, average list entry length, average table entry length, number of links, images, tables, or markup language tags, headlines of tables or lists, or a link associated with an image, list, table, or headline in the electronic document.

18. The computer storage medium of claim 12, wherein filtering the plurality of terms into a first set of terms and a second set of terms comprises:

parsing text in the product page into the plurality of terms;

for each of the plurality of terms, determining a type of the term; and

identifying the term as belonging in the first set of terms or the second set of terms based on the type of the term.

19. The computer storage medium of claim 12, wherein identifying each term of the first set in terms as being associated with a product name or not being associated with the product name comprises:

for each term in the first set of terms, determining one or more features associated with the term and a noun phrase that includes the term; and

determining, with the name recognition model, whether the term is associated with the product name or not associated with the product name based on the one or more features.

20. The computer storage medium of claim 19, wherein determining one or more features associated with the term comprises:

parsing text in the product page into one or more segments;

for each segment, determining a type of the segment; and

associating the type of the segment with the term.

21. A system of one or more computers configured to perform operations comprising:

accessing a set of related electronic documents;

identifying a product page associated with the set of related electronic documents using a page recognition model, the page recognition model generated based on a first machine learning algorithm, and the product page comprising a plurality of terms;

filtering the plurality of terms into a first set of terms and a second set of terms, the first set of terms and the second set of terms including different terms of the plurality of terms, each term in the first set of terms identified as potentially being associated with a product name, and each term in the second set of terms identified as not being associated with a product name; and

identifying each term in the first set of terms as being associated with a product name or not being associated with a product name with a name recognition model, the name recognition model generated based on a second machine learning algorithm.

22. The system of claim 21, wherein identifying a product page associated with the set of related electronic documents using a page recognition model comprises generating the page recognition model by:

accessing an electronic document;

determining features associated with the electronic document;

receiving input specifying a label for the electronic document; and

generating the page recognition model using the labeled electronic document and the features as a training set.

23. The system of claim 21, wherein identifying each term in the first set of terms as being associated with a product name or not being associated with a product name with a name recognition model comprises generating the name recognition model by:

extracting one or more terms from an electronic document;

determining features associated with the one or more terms;

receiving input specifying a label for the one or more terms; and

generating the name recognition model using the labeled terms and the features as a training set.

24. The system of claim 21, wherein identifying a product page comprises:

accessing an electronic document associated with the set of related electronic documents;

determining features associated with the electronic document; and

analyzing the features to determine whether the electronic document is a product page using the page recognition model.

25. The system of claim 24, wherein determining features associated with the electronic document comprises:

analyzing a uniform resource locator, structure, and content associated with the electronic document.

26. The system of claim 25, wherein analyzing the structure associated with the electronic document comprises:

determining at least one of amount of text, length of the longest text section, length of link texts, average link length, average list entry length, average table entry length, number of links, images, tables, or markup language tags, headlines of tables or lists, or a link associated with an image, list, table, or headline in the electronic document.

27. The system of claim 21, wherein filtering the plurality of terms into a first set of terms and a second set of terms comprises:

parsing text in the product page into the plurality of terms;

for each of the plurality of terms, determining a type of the term; and

identifying the term as belonging in the first set of terms or the second set of terms based on the type of the term.

28. The system of claim 21, wherein identifying each term of the first set in terms as being associated with a product name or not being associated with the product name comprises:

for each term in the first set of terms, determining one or more features associated with the term and a noun phrase that includes the term; and

determining, with the name recognition model, whether the term is associated with the product name or not associated with the product name based on the one or more features.

29. The system of claim 28, wherein determining one or more features associated with the term comprises:

parsing text in the product page into one or more segments;

for each segment, determining a type of the segment; and

associating the type of the segment with the term.