Classification rule generation device, classification rule generation method, classification rule generation program, and recording medium

• US 9,323,839 B2
• Filed: 01/13/2011
• Issued: 04/26/2016
• Est. Priority Date: 01/13/2011
• Status: Active Grant

• Current Assignee: Mitsubishi Electric Corporation
• Original Assignee: Mitsubishi Electric Corporation
• Inventors: Shibata, Hideya, Kato, Mamoru, Kori, Mitsunori
• Primary Examiner(s): Badawi, Sherief
• Assistant Examiner(s): Raab, Christopher J
• Application Number:
  US13/996,040
• Publication Number:
  US 20130275433A1
• Time in Patent Office: 1,930 Days
• Field of Search: None
• US Class Current: 1/1
• CPC Class Codes: G06F 16/35 Clustering; Classification
• CPC Class Codes: G06F 16/355 Class or cluster creation or modification
• CPC Class Codes: G06F 16/93 Document management systems

Abstract

In a document classification device 100, a sample document extraction condition storage unit 160 stores sample document extraction conditions 160-1 set for each of classification categories for extracting partial text according to the classification categories from an input document 301 input by a document input unit 110. A document matching unit 120 matches the input document 301 against the sample document extraction conditions 160-1. Based on a result of matching by the document matching unit 120, a document extraction unit 130 extracts the partial text from the input document 301 according to the classification categories. A learning unit 140 performs predetermined machine learning using as a sample document the partial text extracted by the document extraction unit 120, and thereby generates classification rules 150-1.

Claims

1. A classification rule generation device comprising:

an input circuit that inputs a document as a sample target document;

a storage circuit that stores extraction conditions for extracting partial text which is a portion of the sample target document and which is used for generating classification rules for classifying a classification target document to be classified into one of classification categories, the partial text being extracted from the sample target document according to the classification categories, the extraction conditions being set for each of the classification categories;

a matching circuit that matches the sample target document input by the input circuit against the extraction conditions stored in the storage circuit;

an extraction circuit that performs partial text extraction to extract the partial text from the sample target document according to the classification categories, based on a result of matching by the matching circuit; and

a learning circuit that, when the partial text corresponding to one of the classification categories is extracted by the partial text extraction by the extraction circuit, performs predetermined machine learning using the partial text extracted, and generates the classification rules,

wherein the extraction conditions set for each of the classification categories include a keyword that corresponds to each of the classification categories,

the matching circuit includes a position identification circuit that identifies an existing position of the keyword for each of the classification categories in the sample target document,

the extraction circuit extracts a portion around and including the keyword as the partial text from the sample target document, based on the existing position of the keyword identified by the position identification circuit,

the extraction conditions set for each of the classification categories are set such that type information indicating a type of the keyword is set for at least one of the keywords, and

the extraction circuit, when extracting the partial text corresponding to each of the classification categories from the sample target document, performs the partial text extraction based on the type information indicated by the keyword identified by the position identification circuit.

2. The classification rule generation device according to claim 1,

wherein the extraction circuit, when extracting the partial text corresponding to each of the classification categories from the sample target document, does not extract the partial text corresponding to each of the classification categories from the sample target document when a count of type information indicating a specified type out of the type information of the keyword identified by the position identification circuit is less than a set value.

3. The classification rule generation device according to claim 1,

wherein the extraction circuit, when extracting the partial text corresponding to each of the classification categories from the sample target document, uses the sample target document as a target of extraction to extract the partial text corresponding to each of the classification categories only when the type information of the keyword identified by the position identification circuit includes the type information indicating a specified type.

4. The classification rule generation device according to claim 1,

wherein the extraction circuit, when extracting the partial text corresponding to each of the classification categories from the sample target document, changes a size of the partial text corresponding to each of the classification categories in accordance with the type indicated by the type information of the keyword identified by the position identification circuit.

5. The classification rule generation device according to claim 1,

wherein the extraction conditions set for each of the classification categories include at least one inappropriate keyword that does not correspond to each of the classification categories,

wherein the matching circuit includes a position identification circuit that identifies an existing position of the inappropriate keyword for each of the classification categories in the sample target document, and

wherein the extraction circuit excludes a portion around and including the inappropriate keyword from the sample target document, based on the existing position of the inappropriate keyword identified by the position identification circuit.

6. The classification rule generation device according to claim 5,

wherein the extraction conditions set for each of the classification categories are set such that type information indicating a type of the inappropriate keyword is set for at least one of the inappropriate keyword, and

wherein the extraction circuit, when extracting the partial text corresponding to each of the classification categories from the sample target document, performs the partial text extraction based on the type information indicated by the inappropriate keyword identified by the position identification circuit.

7. The classification rule generation device according to claim 6,

wherein the extraction circuit, when extracting the partial text corresponding to each of the classification categories from the sample target document, does not extract the partial text corresponding to each of the classification categories from the sample target document when a count of type information indicating a specified type out of the type information of the inappropriate keyword identified by the position identification circuit is equal to or more than a set value.

8. The classification rule generation device according to claim 6,

wherein the extraction circuit, when extracting the partial text corresponding to each of the classification categories from the sample target document, uses the sample target document as the target of extraction to extract the partial text corresponding to each of the classification categories only when the type information of the inappropriate keyword identified by the position identification circuit does not coincide with the type information indicating a specified type.

9. The classification rule generation device according to claim 6,

wherein the extraction circuit, when extracting the partial text corresponding to each of the classification categories from the sample target document, changes a size of an exclusion range to be excluded from the sample target document in accordance with the type indicated by the type information of the inappropriate keyword identified by the position identification circuit.

10. The classification rule generation device according to claim 5,

wherein the extraction circuit, when extracting the partial text corresponding to each of the classification categories from the sample target document, does not extract the partial text corresponding to each of the classification categories from the sample target document when a count of the inappropriate keyword identified by the position identification circuit is equal to or more than a set value.

11. The classification rule generation device according to claim 1,

wherein the extraction circuit, when a plurality of the partial text having an identical portion are extracted for each of the classification categories, integrates the plurality of the partial text into one piece of the partial text.

12. The classification rule generation device according to claim 1,

wherein the extraction circuit, when extracting the partial text corresponding to each of the classification categories from the sample target document, does not extract the partial text corresponding to each of the classification categories from the sample target document when a count of the keyword identified by the position identification circuit is less than a set value.

13. The classification rule generation device according to claim 1,

wherein the extraction circuit, when extracting the partial text corresponding to each of the classification categories from the sample target document, uses a predetermined range as the target of partial text extraction when a count of the keyword identified by the position identification circuit is equal to or more than a predetermined count within the predetermined range.

14. The classification rule generation device according to claim 1,

wherein the extraction conditions set for each of the classification categories include at least one of an appropriate keyword that is appropriate for each of the classification categories and an inappropriate keyword that is not appropriate for each of the classification categories,

wherein the matching circuit includes a position identification circuit that identifies existing positions of the appropriate keyword and the inappropriate keyword for each of the classification categories in the sample target document, and

wherein the extraction circuit extracts an area including only the appropriate keyword as the partial text from the sample target document, based on the existing positions of the appropriate keyword and the inappropriate keyword identified by the position identification circuit.

15. The classification rule generation device according to claim 14,

wherein the extraction conditions set for each of the classification categories are set such that type information indicating a type of the appropriate keyword is set for at least one of the appropriate keyword, and type information indicating a type of the inappropriate keyword is set for at least one of the inappropriate keyword, and

wherein the extraction circuit, when extracting the partial text corresponding to each of the classification categories from the sample target document, uses the sample target document as the target of extraction to extract the partial text corresponding to each of the classification categories when the type information of the appropriate keyword identified by the position identification circuit coincides with the type information indicating a specified type, and the type information of the inappropriate keyword identified by the position identification circuit does not coincide with the type information indicating a specified type.

16. The classification rule generation device according to claim 1,

wherein the input circuit inputs an electronic mail message as the sample target document,

wherein the extraction conditions of at least one of the classification categories out of the extraction conditions set for each of the classification categories include a specified mail address included in a header of the electronic mail message,

wherein the matching circuit matches the sample target document against the mail address using the extraction conditions, and

wherein the extraction circuit controls the partial text extraction in accordance with a result of matching using the mail address by the matching circuit.

17. The classification rule generation device according to claim 1,

wherein the input circuit inputs a web page as the sample target document,

wherein the extraction conditions of at least one of the classification categories out of the extraction conditions set for each of the classification categories include a uniform resource locator (URL),

wherein the matching circuit matches the sample target document against the URL using the extraction conditions, and

wherein the extraction circuit controls the partial text extraction in accordance with a result of matching using the URL by the matching circuit.

18. The classification rule generation device according to claim 1, further comprising:

a division circuit that divides the classification target document according to an average size of the partial text extracted by the extraction circuit.

19. A classification rule generation device comprising:

an input circuit that inputs a document as a sample target document;

a storage circuit that stores extraction conditions for extracting partial text which is a portion of the sample target document and which is used for generating classification rules for classifying a classification target document to be classified into one of plural classification categories, the partial text being extracted from the sample target document according to the plural classification categories, the extraction conditions being set for each of the plural classification categories;

a matching circuit that matches the sample target document input by the input circuit against the extraction conditions stored in the storage circuit

an extraction circuit that performs partial text extraction to extract the partial text from the sample target document according to the plural classification categories, based on a result of matching by the matching circuit;

a learning circuit that, when the partial text corresponding to one of the plural classification categories is extracted by the partial text extraction by the extraction circuit, performs predetermined machine learning using the partial text extracted, and generates the classification rules,

wherein the extraction conditions set for each of the plural classification categories include a keyword that corresponds to each of the plural classification categories,

the matching circuit includes a position identification circuit that identifies an existing position of the keyword for each of the plural classification categories in the sample target document,

the extraction conditions set for each of the plural classification categories are set such that type information indicating a type of the appropriate keyword is set for at least one of the appropriate keywords, and

the extraction circuit, when extracting the partial text corresponding to each of the plural classification categories from the sample target document, performs the partial text extraction based on the type information indicated by the keyword identified by the position identification circuit.

20. The classification rule generation device according to claim 19,

wherein the extraction circuit extracts a portion around and including the keyword as the partial text from the sample target document, based on the existing position of the keyword identified by the position identification circuit.

21. A classification rule generation method that is executed by a classification rule generation device including an input circuit, a storage circuit, a matching circuit, an extraction circuit, and a learning circuit, the classification rule generation method comprising:

using the input circuit, inputting a sample target document;

using the storage circuit, storing extraction conditions for extracting partial text which is a portion of the sample target document and which is used for generating classification rules for classifying a classification target document to be classified into one of classification categories, the partial text being extracted from the sample target document according to the classification categories, the extraction conditions being set for each of the classification categories;

by the matching circuit, matching the sample target document that is input by the input circuit against the extraction conditions stored in the storage circuit;

by the extraction circuit, performing partial text extraction to extract the partial text from the sample target document according to the classification categories;

by the learning circuit, when the partial text corresponding to one of the classification categories is extracted by the partial text extraction by the extraction circuit, performing predetermined machine learning using the partial text extracted, and generating the classification rules,

by a position identification circuit, identifying an existing position of the keyword for each of the classification categories in the sample target document; and

by the extraction circuit, extracting a portion around and including the keyword as the partial text from the sample target document, based on the existing position of the keyword identified by the position identification circuit,

the extraction conditions set for each of the classification categories including a keyword that corresponds to each of the classification categories,

the extraction conditions set for each of the classification categories are set such that type information indicating a type of the keyword is set for at least one of the keywords, and

the extraction circuit, when extracting the partial text corresponding to each of the classification categories from the sample target document, performs the partial text extraction based on the type information indicated by the keyword identified by the position identification circuit.

22. A non-transitory computer storage medium having instructions stored therein that when executed by processing circuitry perform an information processing method according to claim 21.