APPARATUS, METHOD AND COMPUTER-ACCESSIBLE MEDIUM FOR EXPLAINING CLASSIFICATIONS OF DOCUMENTS

• US 20140229164A1
• Filed: 02/23/2012
• Published: 08/14/2014
• Est. Priority Date: 02/23/2011
• Status: Active Grant

• Current Assignee: New York University
• Original Assignee: New York University
• Inventors: Provost, Foster, Martens, David
• Granted Patent: US 9,836,455 B2
• Time in Patent Office: Days
• US Class Current: 704/9
• CPC Class Codes: G06F 40/169 Annotation, e.g. comment data or footnotes
• CPC Class Codes: G06F 40/279 Recognition of textual entities
• CPC Class Codes: G06F 40/40 Processing or translation of natural language natural language analysis G06F40/20; semantic analysis G06F40/30

Abstract

Classification of collections of items such as words, which are called “document classification,” and more specifically explaining a classification of a document, such as a web-page or website. This can include exemplary procedure, system and/or computer-accessible medium to find explanations, as well as a framework to assess the procedure'"'"'s performance. An explanation is defined as a set of words (e.g., terms, more generally) such that removing words within this set from the document changes the predicted class from the class of interest. The exemplary procedure system and/or computer-accessible medium can include a classification of web pages as containing adult content, e.g., to allow advertising on safe web pages only. The explanations can be concise and document-specific, and provide insight into the reasons for the classification decisions, into the workings of the classification models, and into the business application itself. Other exemplary aspects describe how explaining documents'"'"' classifications can assist in improving the data quality and model performance.

Claims

1. A non-transitory computer readable medium including instructions thereon that are accessible by a hardware processing arrangement, wherein, when the processing arrangement executes the instructions, the processing arrangement is configured to at least generate information associated with a classification of at least one document, comprising:

identifying at least first characteristic of the at least one document;

obtaining at least one second classification of the at least one document after removing the at least one first characteristic of the at least one document; and

generating the information associated with the classification of the at least one document based on the at least one second classification.

2. The non-transitory computer readable medium of claim 1, wherein the at least one first characteristic includes at least one word.

3. The non-transitory computer readable medium of claim 1, wherein the at least one characteristic includes a combination of words.

4. The non-transitory computer readable medium of claim 1, wherein the at least one characteristic includes at least one word, and the processing arrangement is further configured to iteratively obtain the at least one second classification of the at least one document after removing each word in the at least one document.

5. The non-transitory computer readable medium of claim 1, wherein the at least one characteristic includes at least one word, and the processing arrangement is further configured to iteratively obtain the at least one second classification of the at least one document after removing each word and every combination of words in the at least one document.

6. The non-transitory computer readable medium of claim 1, wherein the processing arrangement is further configured to iteratively obtain the at least one second classification of the document after removing the at least one characteristic of the document until the at least one first classification and the at least one second classification are different.

7. The non-transitory computer readable medium of claim 5, wherein the processing arrangement is further configured to omit at least some of the iterations of obtaining the at least one second classification for at least some words or at least some combination of words.

8. The non-transitory computer readable medium of claim 7, wherein the processing arrangement is further configured to omit at least some of the iterations based on at least one of a pruning heuristic search or a hill climbing search.

9. The non-transitory computer readable medium of claim 1, wherein the information includes a minimum-size explanation.

10. The non-transitory computer readable medium of claim 1, wherein the information includes a plurality of minimum explanations.

11. A non-transitory computer readable medium including instructions thereon that are accessible by a hardware processing arrangement, wherein, when the processing arrangement executes the instructions, the processing arrangement is configured to at least generate information associated with at least one classification of a collection, comprising:

identifying at least first characteristic of the collection;

obtaining at least one second classification of the collection after removing the at least one first characteristic of the collection; and

generating the information associated with the classification of the collection based on the at least one second classification.

12. The non-transitory computer readable medium of claim 11, wherein the information includes at least one of an explanation or a hyper-explanation of the at least one first classification of the collection, and wherein the at least one first classification is one of a plurality of classifications.

13. The non-transitory computer readable medium of claim 12, wherein the at least one of an explanation or a hyper-explanation is absent evidence indicating any of the first and second classifications.

14. The non-transitory computer readable medium of claim 13, wherein the at least one of the explanation or the hyper-explanation includes an indication of insufficient vocabulary.

15. The non-transitory computer readable medium of claim 12, wherein the at least one of the explanation or the hyper-explanation includes evidence exclusively indicating at least one of a negative classification or a default classification.

16. The non-transitory computer readable medium of claim 15, wherein the at least one of the explanation or the hyper-explanation is absent evidence of a positive classification.

17. The non-transitory computer readable medium of claim 12, wherein the at least one of the explanation or the hyper-explanation includes evidence exclusively indicating a positive classification.

18. The non-transitory computer readable medium of claim 17, wherein the at least one of the explanation or the hyper-explanation is absent evidence indicating at least one of a negative classification or a default classification.

19. The non-transitory computer readable medium of claim 12, wherein the at least one of the explanation or the hyper-explanation includes evidence indicating a default classification.

20. The non-transitory computer readable medium of claim 12, wherein the at least one of the explanation or the hyper-explanation includes an incorrect prior classification.

21. The non-transitory computer readable medium of claim 12, wherein at least one sets of training data associated with a classifier facilitate generating the at least one of the explanation or the hyper-explanation.

22. The non-transitory computer readable medium of claim 21, wherein the at least one sets of training data includes a set of nearest neighbors that facilitates generating the at least one of the explanation or the hyper-explanation.

23-32. -32. (canceled)

33. A method for generating information associated with at least one classification of a collection, comprising:

identifying, with a processing arrangement, at least first characteristic of the collection;

obtaining at least one second classification of the collection after removing the at least one first characteristic of the collection; and

generating the information associated with the classification of the collection based on the at least one second classification.

67. The method of claim 33, wherein the collection includes at least one document.

34-54. -54. (canceled)

55. A system configured to at least generate information associated with at least one classification of a collection, comprising:

a processing arrangement configured to;

identify at least first characteristic of the collection;

obtain at least one second classification of the collection after removing the at least one first characteristic of the collection; and

generate the information associated with the classification of the collection based on the at least one second classification.

68. The system of claim 55, wherein the collection includes at least one document.

56-66. -66. (canceled)