Feedback system

• US 8,812,419 B1
• Filed: 06/10/2011
• Issued: 08/19/2014
• Est. Priority Date: 06/12/2010
• Status: Active Grant

• Current Assignee: Google LLC (Alphabet Inc.)
• Original Assignee: Google Inc. (Alphabet Inc.)
• Inventors: Teller, Eric, King, Martin T., Mannby, Claes-Fredrik, Furr, Kevin Anthony
• Primary Examiner(s): Gaffin, Jeffrey A
• Assistant Examiner(s): Kim, David H
• Application Number:
  US13/158,305
• Time in Patent Office: 1,166 Days
• Field of Search: 706/12
• US Class Current: 706/46
• CPC Class Codes: G06Q 10/10 Office automation; Time management

Abstract

Exemplary methods and systems are disclosed that apply feedback-based machine learning in order to more intelligently use context information to automate certain actions. An exemplary method involves: determining a first context based on a combination of two or more context signals, using a context-to-action mapping to determine a first action that corresponds to the first context, initiating the first action, after initiating the first action, receiving a data signal corresponding to a user-action, analyzing the data signal to determine a feedback relationship between the user-action and the first action, and based at least in part on the feedback relationship, updating the context-to-action mapping.

Claims

1. A computer-implemented method comprising:

determining a first context signal corresponding to a user-account;

determining a second context signal corresponding to the user-account;

determining, at a computing device, a first context of the user-account based on a combination of two or more context signals, the two or more context signals including the first context signal and the second context signal, wherein the first context comprises a data-based description or characterization of an environment or state that is associated with the user-account;

using a context-to-action mapping of the user-account to determine a first action that corresponds to the first context, wherein the context-to-action mapping maps a plurality of contexts to a plurality of corresponding actions such that the first context maps to the first action;

initiating the first action;

after initiating the first action, receiving a data signal corresponding to a user-action;

analyzing the data signal to determine a feedback relationship between the user-action and the first action; and

based at least in part on the feedback relationship, updating the context-to-action mapping of the user-account.

2. The method of claim 1, wherein the context-to-action mapping indicates a conditional probability that the first action is correct when the second context signal is detected, given that the first context signal is also detected.

3. The method of claim 2, further comprising making a determination to initiate the first action, wherein the determination is based at least in part on the conditional probability.

4. The method of claim 2, wherein updating the context-to-action mapping comprises updating the conditional probability.

5. The method of claim 1, wherein analyzing the data signal to determine the feedback relationship between the user-action and the first action comprises carrying out a machine-learning process based at least in part on the user-action and the first action.

6. The method of claim 1, wherein analyzing the data signal to determine the feedback relationship comprises determining that the user-action implicitly confirms that the first action was correct in the first context.

7. The method of claim 1, wherein analyzing the data signal to determine the feedback relationship comprises determining that the user-action implicitly indicates that the first action was incorrect in the first context.

8. The method of claim 1, wherein analyzing the data signal to determine the feedback relationship comprises determining that the user-action implicitly indicates that the first action is less likely to be correct in the first context.

9. The method of claim 8, wherein the context-to-action mapping includes a likelihood indication that indicates a likelihood that the first action is correct in the first context, and wherein updating the context-to-action mapping comprises adjusting the likelihood indication to indicate a reduced likelihood that the first action is correct in the first context.

10. The method of claim 1, wherein analyzing the data signal to determine the feedback relationship comprises determining that the user-action implicitly indicates that the first action is more likely to be correct in the first context.

11. The method of claim 10, wherein the context-to-action mapping includes a likelihood indication that indicates a likelihood that the first action is correct in the first context, and wherein updating the context-to-action mapping comprises adjusting the likelihood indication to indicate an increased likelihood that the first action is correct in the first context.

12. The method of claim 1, wherein the user-action comprises an action that is dependent, at least in part, on the first action, the method further comprising:

determining, based at least in part on the feedback relationship, that the first action is correct in the first context or is more likely to be correct in the first context.

13. The method of claim 1, wherein the user-action comprises an action that at least in part cancels the first action, the method further comprising:

determining, based at least in part on the feedback relationship, that the first action is not correct in the first context or is less likely to be correct in the first context.

14. The method of claim 1, wherein the first action comprises an action that prompts a user-interaction, wherein the user-action comprises a lack of the user-interaction, the method further comprising:

determining, based at least in part on the feedback relationship, that the first action is not correct in the first context or is less likely to be correct in the first context.

15. The method of claim 1, wherein the first action comprises an action that prompts a user-interaction, wherein the user-action comprises an occurrence of the user-interaction, the method further comprising:

determining, based at least in part on the feedback relationship, that the first action is correct in the first context or is more likely to be correct in the first context.

16. The method of claim 1, wherein the two or more context signals comprise two or more of the following:

(a) a current time, (b) a current date, (c) a current day of the week, (d) a current month, (e) a current season, (f) a time of a future event or future context, (g) a date of a future event or future context, (h) a day of the week of a future event or future context, (i) a month of a future event or future user-context, (j) a season of a future event or future context, (k) a time of a past event or past context, (l) a date of a past event or past context, (m) a day of the week of a past event or past context, (n) a month of a past event or past context, (o) a season of a past event or past context, (p) ambient temperature, (q) a current, future, or past weather forecast at a current location, (r) a current, future, or past weather forecast at a location of a planned event, (s) a current, future, or past weather forecast at or near a location of a previous event, (t) information on a calendar associated with the user-profile, such as information regarding events or statuses of a user or a user'"'"'s friends, (t) information accessible via a user'"'"'s social networking account, such as information relating a user'"'"'s status, statuses of a user'"'"'s friends in a social network group, and/or communications between the user and the users friends, (u) noise level or any recognizable sounds detected by a device, (v) items that are currently detected by a device, (w) items that have been detected in the past by the device, (x) items that other devices associated with a device are currently in communication with or have communicated with in the past, (y) information derived from cross-referencing any two or more of;

information on a user'"'"'s calendar, information available via a user'"'"'s social networking account, and/or other context signals or sources of context information, (z) health statistics or characterizations of a user'"'"'s current health, and (aa) a user'"'"'s recent context as determined from sensors on or near the user and/or other sources of context information.

17. The method of claim 1, wherein the first action comprises a home-automation action, and wherein initiating the first action comprises sending an instruction message to at least one home electronic device.

18. The method of claim 17, wherein the at least one home electronic device comprises at least one of the following:

(a) a home appliance, (b) a refrigerator, (c) a microwave, (d) a dishwasher, (e) a freezer, (f) a washer, (g) a dryer, (h) a combination washer and dryer, (i) a television, (j) a home stereo unit, (k) a laptop computer, (l) a desktop computer, (m) a tablet computer, (n) a smartphone, (o) a garage door opener, (p) a home security system, (q) a home-automation control system, (r) a heating appliance, (s) a cooling appliance, and (t) a cooking appliance.

19. The method of claim 17, wherein the first action comprises an action that adjusts a user-interface of the at least one home-electronic device.

20. The method of claim 1, wherein the first action comprises at least one calendar action.

21. The method of claim 20, wherein the at least one calendar action comprises at least one of the following:

(a) creating a calendar entry, (b) deleting an existing calendar entry, (c) updating an existing calendar entry, (d) setting a reminder for a calendar entry, (e) creating a meeting entry, (f) deleting an existing meeting entry, (g) updating an existing meeting entry, (h) sending one or more invitations for a meeting entry, (i) withdrawing one or more invitations for a meeting entry, (j) setting a reminder for a meeting entry, (k) creating a task entry, (l) deleting a task entry, (m) updating a task entry, and (n) setting a reminder for a task entry.

22. The method of claim 1, wherein the first action comprises an adjustment to a user-interface of the computing device.

23. The method of claim 1, wherein the first action comprises a non-content action.

24. A computer-implemented method comprising:

detecting at least one context signal;

determining, at a computing device, a first context based on the at least one context signal;

using a context-to-action mapping to determine a first action that corresponds to the first context, wherein the context-to-action mapping maps a plurality of contexts to a plurality of corresponding actions such that the at least one context signal maps to the first action;

initiating the first action;

after initiating the first action, receiving a data signal corresponding to a user-action;

analyzing the data signal to determine a feedback relationship between the user-action and the first action, wherein the feedback relationship indicates a desirability difference between the initiation of the first action and one or more previous initiations of the first action in a substantially same context;

responsive to the feedback, determining one or more context signals that have a likelihood of reconciling the desirability difference between the initiation of the first action and one or more previous initiations of the first action in a substantially same context; and

updating the context-to-action mapping of the user-account to indicate that the first context should be determined based on both;

(a) the at least one context signal and (b) the one or more context signals that have a likelihood of reconciling the desirability difference.

25. The method of claim 24, wherein determining the one or more context signals that have a likelihood of reconciling the desirability difference comprises analyzing data from a historical context-to-action database to determine the one or more context signals that have a likelihood of reconciling the desirability difference between the initiation of the first action and one or more previous initiations of the first action in a substantially same context.

26. The method of claim 24, wherein determining the one or more context signals that have a likelihood of reconciling the desirability difference comprises initiating a prompt for input data to determine the one or more context signals that have a likelihood of reconciling the desirability difference between the initiation of the first action and one or more previous initiations of the first action in a substantially same context.

27. A system comprising:

a non-transitory computer-readable medium; and

program instructions stored on the non-transitory computer-readable medium and executable by at least one processor to;

determine a first context of a user-account based on a combination of two or more context signals corresponding to the user-account, the two or more context signals including a first context signal and a second context signal, wherein the first context comprises a data-based description or characterization of an environment or state that is associated with the user-account;

use a context-to-action mapping of the user-account to determine a first action that corresponds to the first context, wherein the context-to-action mapping maps a plurality of contexts to a plurality of corresponding actions such that the first context maps to the first action;

initiate the first action;

after initiating the first action, receive a data signal corresponding to a user-action;

analyze the data signal to determine a feedback relationship between the user-action and the first action; and

based at least in part on the feedback relationship, update the context-to-action mapping of the user-account.

28. The system of claim 27, wherein the context-to-action mapping indicates a conditional probability that the first action is correct when the second context signal is detected, given that the first context signal is also detected.

29. The system of claim 27, further comprising program instructions stored on the non-transitory computer-readable medium and executable by at least one processor to condition the initiation of the first action on a determination as to whether or not to initiate the first action, wherein the determination is based at least in part on the conditional probability.

30. The system of claim 27, wherein the update of the context-to-action mapping includes an adjustment of the conditional probability.

31. The system of claim 27, further comprising program instructions stored on the non-transitory computer-readable medium and executable by at least one processor to carry out a machine-learning process to determine a feedback relationship between the user-action and the first action, wherein the machine-learning process is based at least in part on the user-action and the first action.

32. The system of claim 27, wherein the feedback relationship comprises an implicit indication as to whether or not the user-action is correct in the first context.

33. The system of claim 27, wherein the feedback relationship comprises an implicit indication as to whether the user-action is more or less likely to be correct in the first context.

34. The system of claim 33:

wherein the context-to-action mapping includes a likelihood indication that indicates a likelihood that the first action is correct in the first context;

wherein, if the implicit indication indicates that the user-action is more likely to be correct, then the update of the context-to-action mapping includes an adjustment of the likelihood indication to indicate an increased likelihood that the first action is correct in the first context; and

wherein, if the implicit indication indicates that the user-action is less likely to be correct, then the update of the context-to-action mapping includes an adjustment of the likelihood indication to indicate a reduced likelihood that the first action is correct in the first context.

35. The system of claim 27, wherein the two or more context signals comprise two or more of the following:

(a) a current time, (b) a current date, (c) a current day of the week, (d) a current month, (e) a current season, (f) a time of a future event or future context, (g) a date of a future event or future context, (h) a day of the week of a future event or future context, (i) a month of a future event or future user-context, (j) a season of a future event or future context, (k) a time of a past event or past context, (l) a date of a past event or past context, (m) a day of the week of a past event or past context, (n) a month of a past event or past context, (o) a season of a past event or past context, (p) ambient temperature, (q) a current, future, or past weather forecast at a current location, (r) a current, future, or past weather forecast at a location of a planned event, (s) a current, future, or past weather forecast at or near a location of a previous event, (t) information on a calendar associated with the user-profile, such as information regarding events or statuses of a user or a user'"'"'s friends, (t) information accessible via a user'"'"'s social networking account, such as information relating a user'"'"'s status, statuses of a user'"'"'s friends in a social network group, and/or communications between the user and the users friends, (u) noise level or any recognizable sounds detected by a device, (v) items that are currently detected by a device, (w) items that have been detected in the past by the device, (x) items that other devices associated with a device are currently in communication with or have communicated with in the past, (y) information derived from cross-referencing any two or more of;

information on a user'"'"'s calendar, information available via a user'"'"'s social networking account, and/or other context signals or sources of context information, (z) health statistics or characterizations of a user'"'"'s current health, and (aa) a user'"'"'s recent context as determined from sensors on or near the user and/or other sources of context information.

36. The system of claim 27, wherein the first action comprises a home-automation action, and wherein the first action comprises sending an instruction message at least one a home electronic device.

37. The system of claim 36, wherein the at least one home electronic device comprises at least one of the following:

(a) a home appliance, (b) a refrigerator, (c) a microwave, (d) a dishwasher, (e) a freezer, (f) a washer, (g) a dryer, (h) a combination washer and dryer, (i) a television, (j) a home stereo unit, (k) a laptop computer, (l) a desktop computer, (m) a tablet computer, (n) a smartphone, (o) a garage door opener, (p) a home security system, (q) a home-automation control system, (r) a heating appliance, (s) a cooling appliance, and (t) a cooking appliance.

38. The system of claim 27, wherein the first action comprises at least one calendar action.

39. The system of claim 38, wherein the at least one calendar action comprises at least one of the following:

(a) creating a calendar entry, (b) deleting an existing calendar entry, (c) updating an existing calendar entry, (d) setting a reminder for a calendar entry, (e) creating a meeting entry, (f) deleting an existing meeting entry, (g) updating an existing meeting entry, (h) sending one or more invitations for a meeting entry, (i) withdrawing one or more invitations for a meeting entry, (j) setting a reminder for a meeting entry, (k) creating a task entry, (l) deleting a task entry, (m) updating a task entry, and (n) setting a reminder for a task entry.

40. The system of claim 27, wherein the first action comprises an adjustment to a user-interface of the computing device.

41. A system comprising:

a means for determining a first context of a user-account based on a combination of two or more context signals corresponding to the user-account, wherein the first context comprises a data-based description or characterization of an environment or state that is associated with the user-account;

a means for using a context-to-action mapping of the user-account to determine a first action that corresponds to the first context, wherein the context-to-action mapping maps a plurality of contexts to a plurality of corresponding home-automation actions, such that the first context maps to the first action;

a means for initiating the first action;

a means for, after initiating the first action, receiving a data signal corresponding to a user-action;

a means for, analyzing the data signal to determine a feedback relationship between the user-action and the first action; and

a means for, based at least in part on the feedback relationship, updating the context-to-action mapping of the user-account.

42. A computer-implemented method comprising:

determining a first context signal corresponding to a user-account, wherein the first context signal provides information related to an environment or state that is associated with the user-account;

determining a second context signal corresponding to the user-account, wherein the second context signal provides information related to an environment or state that is associated with the user-account;

using a context-to-action mapping of the user-account to determine a first action that corresponds to a combination of the first context signal and the second context signal, wherein the context-to-action mapping maps a plurality of context-signal combinations to a plurality of corresponding actions such that the combination of the first context signal and the second context signal maps to the first action;

initiating the first action;

after initiating the first action, receiving a data signal corresponding to a user-action;

analyzing the data signal to determine a feedback relationship between the user-action and the first action; and

based at least in part on the feedback relationship, updating the context-to-action mapping of the user-account.

43. A computer-implemented method comprising:

detecting at least one context signal;

determining, at a computing device, a first context based on the at least one context signal;

using a context-to-action mapping to determine a first action that corresponds to the first context, wherein the context-to-action mapping maps a plurality of contexts to a plurality of corresponding actions such that the at least one context signal maps to the first action;

initiating the first action;

after initiating the first action, receiving a data signal corresponding to a user-action;

analyzing the data signal to determine a feedback relationship between the user-action and the first action, wherein the feedback relationship indicates a desirability difference between the initiation of the first action and one or more previous initiations of the first action in a substantially same context;

responsive to the feedback, initiating a prompt for input data to determine one or more context signals that have a likelihood of reconciling the desirability difference between the initiation of the first action and one or more previous initiations of the first action in a substantially same context; and

updating the context-to-action mapping to indicate that the first context should be determined based on both;

(a) the at least one context signal and (b) the one or more context signals that have a likelihood of reconciling the desirability difference.