Automated assistant for organizing electronic documents

• US 7,051,277 B2
• Filed: 04/17/1998
• Issued: 05/23/2006
• Est. Priority Date: 04/17/1998
• Status: Expired due to Fees

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Kephart, Jeffrey Owen, White, Steve Richard, Segal, Richard Brian
• Primary Examiner(s): BASHORE, WILLIAM L
• Application Number:
  US09/061,706
• Publication Number:
  US 20010042087A1
• Time in Patent Office: 2,958 Days
• Field of Search: 707/1, 707/3, 707/104, 707/102, 707/5, 715/530, 715/531, 706/20, 706/45, 345/354, 709/219
• US Class Current: 715/229
• CPC Class Codes: G06F 16/353 into predefined classes
• CPC Class Codes: Y10S 707/99931 Database or file accessing
• CPC Class Codes: Y10S 707/99933 Query processing, i.e. searching
• CPC Class Codes: Y10S 707/99945 Object-oriented database structure processing

Abstract

A method of assisting a user with the task of categorizing a received electronic document into a collection includes the steps of classifying the document to obtain one or more most likely categorical labels; displaying, to the user, a representation of the one or more most likely categorical labels; receiving data, from the user, representative of a selected categorical label; and labeling the document within the collection with the selected categorical label. The electronic document can include an electronic mail message, a web page bookmark, an audio file or a video file.

Claims

1. An automated method of assisting a user with the task of categorizing electronic documents into a collection, comprising the steps of:

classifying, with a classifier, a document to obtain a plurality of most likely categorical labels;

deriving a plurality of categorizations shortcuts from the plurality of most likely categorical labels;

displaying, to the user, a representation of the plurality of most likely categorical labels;

receiving, from the user, a selection of one or more of the most likely categorical labels representative of the document to be categorized within a collection;

labeling the document within the collection with one or more of the selected categorical labels; and

incrementally retraining the classifier to adapt to modifications of the collection, wherein the incremental retraining is performed using a lazy strategy for incrementally retraining the classifier.

2. The method of claim 1 wherein the classifying step comprises the step of classifying, upon receipt into data storage, the document to obtain the plurality of most likely categorical labels.

3. The method of claim 1 wherein the deriving step comprises the step of deriving, upon receipt of the document into data storage, categorization shortcuts from the plurality of most likely categorical labels.

4. The method of claim 1 wherein the deriving step comprises the step of labeling display buttons with the plurality of most likely categorical labels, and the displaying step comprises the step of displaying the labeled display buttons with the document.

5. The method of claim 1 wherein the deriving step comprises the step of creating an ordered set of the plurality of most likely categorical labels, and the displaying step comprises the step of displaying with the document the ordered set prepended to a standard ordered set of other categorical labels.

6. The method of claim 1 wherein the classifying step occurs substantially simultaneously with the displaying step.

7. The method of claim 1 wherein the classifying step comprises the step of classifying upon invocation by the user, the document to obtain the plurality of most likely categorical labels.

8. The method of claim 7 wherein the invocation comprises a selection by the user of a classify button.

9. The method of claim 1 wherein the labeling step comprises the step of storing the document in folders or locations of the collection corresponding to the one or more selected categorical labels.

10. The method of claim 9, wherein the classifying step comprises the steps of:

tokenizing the document into different tokens;

tallying a number of occurrences of each token in the document;

computing, for each folder, a token weight of each token;

comparing, for each token, the number of occurrences and the token weights;

creating a similarity score in response to the comparing step; and

identifying a subset of folders for which the similarity score is highest.

11. The method of claim 10 further comprising the step of removing, from the identified subset, all folders from which the similarity score is lower than a default or specified threshold.

12. The method of claim 10, wherein the computing step comprises the step of computing the token counts of each token in each of the folders.

13. The method of claim 10 wherein the tokenizing step comprises the steps of:

separately tokenizing different portions of the document; and

labeling the tokens according to the different portions.

14. The method of claim 1 further comprising the step of displaying a standard list of all categorical labels, wherein the receiving step comprises the step of receiving, from the user, data representative of one or more selected categorical labels from either the plurality of displayed categorization shortcuts or the standard list.

15. The method of claim 1 wherein the retraining step comprises the step of retraining the classifier in response to the labeling step.

16. The method of claim 1 wherein the labeling step comprises the step of storing the document in folders or locations of the collection corresponding to the one or more selected categorical labels and the retraining step comprises the steps of:

receiving, from the user, addition data representative of an addition of a document into a tofolder; and

retraining the classifier in response to the addition data.

17. The method of claim 16 wherein the retraining step comprises the step of assigning, in the classifier, the added document to the tofolder.

18. The method of claim 17 further comprising the step of identifying excluded folders to be excluded from retraining and wherein the retraining step comprises the step of assigning, in the classifier, the added document when the tofolder is not one of the identified excluded folders.

19. The method of claim 17 wherein the classifying step comprises the steps of:

tokenizing the document into different tokens;

tallying a number of occurrences of each token in the document;

retrieving, for each folder, a tokencount of each token;

computing, for each folder, a token weight of each token;

comparing, for each token, the number of occurrences and the token weights;

creating a similarity score in response to the comparing step; and

identifying a subset of folders for which the similarity score is highest, and wherein the assigning step comprises the step of adding the number of occurrences of each token to the tokencount of the tofolder.

20. The method of claim 19 wherein the electronic document is web page and the collection is a collection of bookmarks, the method further comprising the step of storing, for each web page, a pagetokencount matching the tallied number of occurrences of each token.

21. The method of claim 1 wherein the labeling step comprises the step of storing the document in folders or locations of the collection corresponding to the one or more selected categorical labels and the retraining step comprises the steps of:

receiving, from the user, deletion data representative of a removal of a document from a from folder; and

retraining the classifier in response to the deletion data.

22. The method of claim 21 wherein the retraining step comprises the step of unassigning, in the classifier, the removed document from the from folder in which it was categorized.

23. The method of claim 22 further comprising the step of identifying excluded folders to be excluded from retraining and wherein the retraining step comprises the step of unassigning, in the classifier, the removed document when the from folder is not one of the identified excluded folders.

24. The method of claim 22 wherein the classifying step comprises the steps of:

tokenizing the document into different tokens;

tallying a number of occurrences of each token in the document;

retrieving, for each folder, a tokencount of each token;

computing, for each folder, a token weight of each token;

comparing, for each token, the number of occurrences and the token weights;

creating a similarity score in response to the comparing step; and

identifying a subset of folders for which the similarity score is highest, and wherein the unassigning step comprises the step of subtracting the number of occurrences of each token to the tokencount of the tofolder.

25. The method of claim 24 wherein the electronic document is a web page and the collection is a collection of bookmarks, the method further comprising the step of storing, for each web page, a pagetokencount matching the tallied number of occurrences of each token, wherein the unassigning step comprises the step of subtracting the pagetokencount from the tokencount of the fromfolder.

26. The method of claim 1 wherein the labeling step comprises the step of storing the document in folders or locations of the collection corresponding to the one or more selected categorical labels and the retraining step comprises the steps of:

receiving, from the user, move data representative of a movement of a document from a source folder to a destination folder; and

retraining the classifier in response to the move data.

27. The method of claim 26 wherein the retraining step comprises the steps of:

unassigning, the classifier, the moved document from the source folder in which it was categorized; and

assigning, in the classifier, the moved document to the destination folder.

28. The method of claim 27 further comprising the step of identifying excluded folders to be excluded from retraining and wherein the retraining step comprises the steps of:

unassigning, in the classifier, the moved document when the source folder is not one of the identified excluded folders; and

assigning, in the classifier, the moved document when the destination folder is not one of the identified excluded folders.

29. The method of claim 27 wherein the classifying steps comprises the steps of:

tokenizing the document into different tokens;

tallying a number of occurrences of each token in the document;

retrieving, for each folder, a tokencount of each token;

computing, for each folder, a token weight of each token;

comparing, for each token, the number of occurrences and the token weights;

creating a similarity score in response to the comparing step; and

identifying a subset of folders for which the similarity score is highest, and wherein the unassigning step comprises the step of subtracting the number of occurrences of each token from the tokencount of the source folder, and the assigning step comprises the step of adding the number of occurrences of each token to the tokencount of the destination folder.

30. The method of claim 29 wherein the electronic document is a web page and the collection is a collection of bookmarks, the method further comprising the step of storing, for each web page, a pagetokencount matching the tallied number of occurrences of each token, wherein the unassigning step comprises the step of subtracting the pagetokencount from the tokencount of the fromfolder.

31. The method of claim 1 wherein the retraining step occurs immediately after a collection modification.

32. The method of claim 1, further comprising the step of training the classifier from scratch with a pre-existing collection of categorized documents.

33. The method of claim 32 wherein the labeling step comprises the step of storing the document in folders or locations of the collection corresponding to the one or more selected categorical labels and the training step comprises the step of assigning, in the classifier, each of the pre-existing documents to a folder in which it is categorized.

34. The method of claim 33 wherein the classifying step comprises the steps of:

tokenizing the document into different tokens;

tallying a number of occurrences of each token in the document;

retrieving, for each folder, a tokencount of each token;

computing, for each folder, a token weight of each token;

comparing, for each token, the number of occurrences and the token weight;

creating a similarity score in response to the comparing step; and

identifying a subset of folders for which the similarity score is highest, and wherein the assigning step comprises the step of adding the number of occurrences of each token to the tokencount of the tofolder.

35. The method of claim 33 further comprising the step of identifying excluded folders to be excluded from training and wherein the training step comprises the step of assigning, in the classifier, each of the pre-existing documents, except those in the identified excluded folders.

36. The method of claim 1 wherein the labeling step comprises the step of storing the document in folders or locations of the collection corresponding to the one or more selected categorical labels and the re-training step comprises the steps of:

determining a time of a last step of retraining; and

retraining the classifier on each folder which was modified after the determined time.

37. The method of claim 1 wherein labeling step comprises the step of storing the document in folders or locations of the collection corresponding to the one or more selected categorical labels, the method further comprising the step of training the classifier from scratch with a pre-existing collection of categorized documents, wherein the retraining step comprises the steps of:

determining a time of the step of training or a last step of re-training; and

retraining the classifier on each folder which was modified after the determined time.

38. The method of claim 1 wherein the classifying step uses the TF-IDF principle.

39. The method of claim 1 wherein the electronic document is an e-mail message.

40. The method of claim 1 wherein the electronic document is web page and the collection is a collection of bookmarks.

41. The method of claim 1 wherein the electronic document is a multimedia document.

42. The method of claim 41 wherein the multimedia document is an image file, a video file or an audio file.

43. The method of claim 42 wherein the multimedia document further includes text.

44. The method of claim 41 wherein the multimedia document combines any combination of text, an image file, a video file and an audio file.

45. The method of claim 1 wherein the electronic document comprises data sets that are not viewable in their entirety, but can be categorized in response to some presentation to the user.

46. The method of claim 1 wherein:

the step of displaying most likely categorical labels to the user further comprises displaying through a graphical user interface a plurality of virtual category buttons, each labeled with one of the categorical labels;

the step of receiving representative data from the user comprises receiving information that one of the category buttons has been selected by the user by clicking on the category button; and

the step of labeling the document comprises virtually moving the document to a virtual file folder unique to the categorical label selected, and occurs upon the clicking of the selected category button without need for any other activity by the user.

47. The automated method of claim 1, wherein the incrementally retraining is performed using a lazy strategy for incrementally retraining the classifier comprises:

deferring retraining of the classifier;

performing bookkeeping operations whenever messages are added to folders, removed from folders, or moved from one folder to another; and

automatically triggering retraining of the classifier by a predetermined criteria, wherein the retraining the classifier by processing any updates that have been deferred.

48. The automatic method of claim 47, wherein the predetermined criteria for automatically performing the retraining step is a fixed amount of time that has elapsed since a last retraining step has been performed.

49. The automatic method of claim 47, wherein the predetermined criteria for automatically performing the retraining step is when a threshold number of documents have been added, deleted, or moved in the collection and any combination thereof.

50. The automatic method of claim 47, wherein the predetermined criteria for automatically performing the retraining step is when the system has reached an idle state or is either updating the classifier or performing bookkeeping operations whenever messages are added to folders, removed from folders, or moved from one folder to another.

51. The automatic method of claim 1, wherein incremental retraining the classifier includes an instant strategy for incrementally retraining the classifier.

52. A program storage device, readable by a machine, tangibly embodying a program of instructions executable by the machine to perform method steps for automatically assisting a user with the task of categorizing an electronic document into a collection, the method comprising the steps of:

classifying, with a classifier, a document to obtain a plurality of most likely categorical labels;

deriving a plurality of categorizations shortcuts from the plurality of most likely categorical labels;

displaying, to the user, a representation of the plurality of most likely categorical labels;

receiving, from the user, a selection of one or more of the most likely categorical labels representative of the document to be categorized within a collection;

labeling the document within the collection with one or more of the selected categorical labels; and

incrementally retraining the classifier to adapt to modifications of the collection, wherein the incremental retraining is performed using a lazy strategy for incrementally retraining the classifier.

53. The program storage device of claim 52 wherein:

the step of displaying most likely categorical labels to the user further comprises displaying through a graphical user interface a plurality of virtual category buttons, each labeled with one of the categorical labels; and

the step of receiving representative data from the user comprises receiving information that one of the category buttons has been selected by the user by clicking on the category button; and

the step of labeling the document comprises virtually moving the document to virtual file folder unique to the categorical label selected, and occurs upon the clicking of the selected category button without need for any other activity by the user.