Learning by observing a user's activity for enhancing the provision of automated services

• US 6,581,050 B1
• Filed: 04/20/1999
• Issued: 06/17/2003
• Est. Priority Date: 04/20/1999
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Horvitz, Eric, Jacobs, Andrew W.
• Primary Examiner(s): STARKS, WILBERT L
• Application Number:
  US09/295,077
• Time in Patent Office: 1,519 Days
• Field of Search: 706/12, 706/45, 345/707, 434/350, 703/27, 705/1, 705/26, 707/5
• US Class Current: 706/45
• CPC Class Codes: G06F 40/216 using statistical methods
• CPC Class Codes: G06F 40/289 Phrasal analysis, e.g. finite state techniques or chunking
• CPC Class Codes: G06Q 10/109 Time management, e.g. calendars, reminders, meetings or time accounting

Abstract

Learning by observing a user'"'"'s activity for enhancing the provision of automated services. In one embodiment, a computer-implemented method first determines a text to analyze. The method then determines the probabilities of a user'"'"'s goals associated with the text based on features observed in the text, related materials, or user actions, and based on the probabilities of a user'"'"'s goals, selects one of the following options: (1) inaction, (2) automatic action, or (3) dialog with users about potential actions to seek a user'"'"'s approval. Upon the method selecting either the (1) automatic action option or the (2) dialog with user about suggested action—the latter also in conjunction with receiving actual user approval—the method performs an action based on the text. In one embodiment, the method waits a predetermined time prior to action, determined by performing a statistical regression as to the predetermined time that should be waited based on a length of each text. Furthermore, in one embodiment the action probability is determined by a text classification system, such that the system is continually trained based on each text inputted thereinto.

Claims

1. A computer-implemented method comprising:

determining a text to analyze;

determining an action probability based on the text;

selecting an option selected from the group essentially consisting of;

(1) inaction, (2) automatic action, and (3) suggested action with user approval, based on the action probability; and

,

upon selecting the option comprising one of the (1) automatic action and the (2) suggested action with user approval, the latter also in conjunction with receiving actual user approval, performing an action based on the text.

2. The method of claim 1, further comprising waiting a predetermined time prior.

3. The method of claim 2, wherein waiting a predetermined time initially comprises determining the predetermined time comprising, for each text received, performing a statistical regression as to the predetermined time that should be waited based on a length of each text.

4. The method of claim 2, wherein waiting a predetermined time comprises waiting a predetermined time between a minimum time and a maximum time.

5. The method of claim 4, wherein at least one of the minimum time and the maximum time is defined by a user.

6. The method of claim 1, wherein the action comprises one of scheduling, forwarding, paging, routing and moving.

7. The method of claim 1, wherein determining an action probability based on the text comprises inputting the text into a text classification system to determine the action probability.

8. The method of claim 7, wherein inputting the text into a text classification system comprises continually training the text classification system based on each text inputted into the text classification system.

9. The method of claim 7, wherein inputting the text into a text classification system comprises initially generating the text classification system.

10. The method of claim 9, wherein initially generating the text classification system comprises:

performing a feature-selection phase; and

,

performing a model-construction phase based on the feature-selection phase.

11. The method of claim 10, wherein performing a feature-selecting phase comprises training the text classification system by inputting a plurality of texts based on which action is to occur and inputting a plurality of texts based on which action is not to occur.

12. The method of claim 10, wherein performing a feature-selection phase comprises seeding the feature-selection phase with domain-specific knowledge.

13. The method of claim 1, wherein selecting an option comprises:

associating a boundary between the (1) and the (2) options with a first probability threshold, and a boundary between the (2) and the (3) options with a second probability threshold;

upon determining that the action probability is less than the first probability threshold, selecting the (1) option;

upon determining that the action probability is greater than the second probability threshold, selecting the (3) option; and

,

otherwise, selecting the (2) option.

14. The method of claim 13, wherein associating a boundary between the (1) and the (2) options with a first probability threshold, and a boundary between the (2) and the (3) options with a second probability threshold comprises utilizing decision theory to determine the first and the second probability thresholds.

15. The method of claim 14, wherein utilizing decision theory comprises utilizing cost/benefit analysis.

16. The method of claim 15, wherein utilizing cost/benefit analysis comprises utilizing the cost/benefit analysis in a contextual manner.

17. A machine-readable medium having processor instructions stored thereon for execution by a processor to cause performance of a method comprising:

determining a text to analyze;

determining an action probability based on the text;

selecting an option selected from the group essentially consisting of;

(1) inaction, (2) automatic action, and (3) suggested action with user approval, based on the action probability; and

,

upon selecting the option comprising one of the (1) automatic action and the (2) suggested action with user approval, the latter also in conjunction with receiving actual user approval, performing an action based on the text.

18. The medium of claim 17, further comprising waiting a predetermined time.

19. The medium of claim 18, wherein waiting a predetermined time initially comprises determining the predetermined time comprising, for each text received, performing a statistical regression as to the predetermined time that should be waited based on a length of each text.

20. The medium of claim 18, wherein waiting a predetermined time comprises waiting a predetermined time between a minimum time and a maximum time.

21. The medium of claim 20, wherein at least one of the minimum time and the maximum time is defined by a user.

22. The medium of claim 17, wherein the action comprises one of scheduling, forwarding, paging, routing and moving.

23. The medium of claim 17, wherein determining a text to analyze comprises determining a message having focus.

24. The medium of claim 17, wherein determining an action probability based on the text comprises inputting the text into a text classification system to determine the action probability.

25. The medium of claim 24, wherein inputting the text into a text classification system comprises continually training the text classification system based on each text inputted into the text classification system.

26. The medium of claim 24, wherein inputting the text into a text classification system comprises initially generating the text classification system.

27. The medium of claim 24, wherein initially generating the text classification system comprises:

performing a feature-selection phase; and

,

performing a model-construction phase based on the feature-selection phase.

28. The medium of claim 27, wherein performing a feature-selecting phase comprises training the text classification system by inputting a plurality of texts based on which action is to occur and inputting a plurality of texts based on which action is not to occur.

29. The medium of claim 27, wherein performing a feature-selection phase comprises seeding the feature-selection phase with domain-specific knowledge.

30. The medium of claim 17, wherein selecting an option comprises:

associating a boundary between the (1) and the (2) options with a first probability threshold, and a boundary between the (2) and the (3) options with a second probability threshold;

upon determining that the action probability is less than the first probability threshold, selecting the (1) option;

upon determining that the action probability is greater than the second probability threshold, selecting the (3) option; and

,

otherwise, selecting the (2) option.

31. The medium of claim 30, wherein associating a boundary between the (1) and the (2) options with a first probability threshold, and a boundary between the (2) and the (3) options with a second probability threshold comprises utilizing decision theory to determine the first and the second probability thresholds.

32. The medium of claim 31, wherein utilizing decision theory comprises utilizing cost/benefit analysis.

33. The medium of claim 32, wherein utilizing cost/benefit analysis comprises utilizing the cost/benefit analysis in a contextual manner.

34. A computerized system comprising:

a program in which a user is able to perform an action; and

,

an action direction program to determine an action probability based on an input;

select an option selected from the group essentially consisting of;

(1) inaction, (2) automatic action, and (3) suggested action with user approval, based on the action probability; and

, upon selecting the option comprising one of the (1) automatic action and the (2) suggested action with user approval, the latter also in conjunction with receiving actual user approval, performing an action within the program based on the input.

35. The system of claim 34, wherein the action direction program initially waits a predetermined time.

36. The system of claim 35, wherein the predetermined time is determined by a statistical regression as to the predetermined time that should be waited based on a length of each text.

37. The system of claim 34, wherein the action direction program continually trains a text classification system based on each text.

38. The system of claim 34, wherein the input of the action direction program comprises a user-provided input, such that that the action direction program is user-invoked.

39. The system of claim 34, wherein the action program comprises a computer program executed by a processor from a computer-readable medium.

40. The system of claim 34, wherein the action direction program comprises a computer program executed by a processor from a computer-readable medium.

41. The system of claim 34, wherein the action comprises one of scheduling, forwarding, paging, routing and moving.