Continuous time bayesian network models for predicting users' presence, activities, and component usage

• US 20050021485A1
• Filed: 06/30/2004
• Published: 01/27/2005
• Est. Priority Date: 06/28/2001
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Horvitz, Eric J., Nodelman, Uri D.
• Granted Patent: US 7,689,521 B2
• Time in Patent Office: Days
• US Class Current: 706/21
• CPC Class Codes: G06Q 10/109 Time management, e.g. calendars, reminders, meetings or time accounting

Abstract

The present invention relates to a system and methodology to facilitate collaboration and communications between entities such as between automated applications, parties to a communication and/or combinations thereof. The systems and methods of the present invention include a service that supports collaboration and communication by learning predictive continuous time Bayesian models that provide forecasts of one or more aspects of a users'"'"' presence and availability. Presence forecasts include a user'"'"'s current or future locations at different levels of location precision and usage of different devices or applications. Availability assessments include inferences about the cost of interrupting a user in different ways and a user'"'"'s current or future access to one or more communication channels. The predictive models are constructed from data collected by considering user activity and proximity from multiple devices, in addition to analysis of the content of users'"'"' calendars, the time of day, and day of week, for example. Various applications are provided that employ the presence and availability information supplied by the models in order to facilitate collaboration and communications between entities.

Claims

1. A prediction system to facilitate communications and collaboration between entities, comprising:

a presence component to determine a user'"'"'s state at a computer; and

an application component associated with the presence component that form a model to generate predictions relating to states of the user and at least one application, the model adapted in accordance with a continuous time Bayesian network.

2. The system of claim 1, the model is employed to predict the user'"'"'s activities and component usage.

3. The system of claim 1, the model predicts an application that is on currently on top or at the focus of the user'"'"'s attention.

4. The system of claim 1, the model employed to determine when the user is expected to be present on a computer.

5. The system of claim 1, the model employed to determine when the user is expected to access the application.

6. The system of claim 5, the application includes at least one of an email application, a calendar application, a text application, a presentation application, a graphics application, a software development application, an audio application, a video application, a spreadsheet application, an Internet application, and a communications application.

7. The system of claim 1, the model employed to determine how much time the user is expected to remain on a computer.

8. The system of claim 1, the model employed to determine how much time before the user is expected to change to a different application.

9. The system of claim 1, further comprising a calendar component that indicates at least one of a weekday or a weekend.

10. The system of claim 1, further comprising discretized a variable keeps track of how long it has been since the user last used the application.

11. The system of claim 1, further comprising a forecasting component that receives one or more queries regarding an identified user'"'"'s presence or availability and generates one or more predictions relating to current or future states of the identified user.

12. The system of claim 11, the forecasting component determines complementary information relating to how long a person is expected to be absent, or how long a person may be unavailable.

13. The system of claim 11, the forecasting component determines presence states that include at least one of a time until a user will arrive at or leave a location, a time until the user will be at a location for at least time t, a time until the user will have access to a device, a time until the user will review an e-mail or other message, a time until the user will finish a conversation in progress, a likelihood the user will attend a meeting, and an expected cost of interruption over time.

14. The system of claim 11, the forecasting component is implemented as at least one of a server, a server farm, a client application, a web service, and an automated application that provides answers to automated systems or authorized people.

15. The system of claim 11, further comprising an event database that logs periods of presence and absence as events in an event log.

16. The system of claim 15, the events are annotated by source devices that are defined by respective capabilities and locations.

17. The system of claim 16, further comprising tagging of events by specific devices, indexed by capabilities to enable a system to forecast a probability distribution over a time until the user will have access to different types of devices.

18. The system of claim 11, the forecasting component predicts a time when a current conversation is likely to end.

19. A computer readable medium having computer readable instructions stored thereon for implementing the components of claim 1.

20. A method to provide presence forecasts, comprising:

determining a presence model to predict a user'"'"'s presence or absence from a computer; and

determining an application model to facilitate determining the user'"'"'s presence or absence from the computer, at least one of the presence model and the application model configured according to a continuous time Bayesian network.

21. The method of claim 20, further comprising determining a total loss for a trajectory.

22. The method of claim 20, further comprising determining a local loss as a measure of how well a particular transition is predicted by a model.

23. The method of claim 20, further comprising determining a loss function as zero at a point of transition and nonnegative for left and right portions of the transition.

24. The method of claim 23, further comprising determining a basic shape of the loss function having independent parameters for the left and right portions of the loss function.

25. The method of claim 24, the loss function is a linear function having the form:

lin<

alpha, beta, t′

>

(t)=alpha(t′

−

t) for t<

t′

beta(t−

t′

) otherwise

where t′

is an actual time of transition, t is a predicted time of transition, and parameters alpha and beta are slopes for linear portions of the loss function.

26. The method of claim 25, further comprising determining an expected loss as an expected value of the loss function given a distribution over when a transition will take place. We can include in the loss a penalty term if the transition is not predicted by the model.

27. The method of claim 26, the expected loss is computed as a function of a relevant intensity parameter q and an actual time of transition t.

29. The method of claim 20, further comprising determining a phase distribution as a distribution over a time to absorption to an absorbing state of a continuous time Markov process.

28. The method of claim 28, the expected loss is computed according to the following formula:

CTBNExpectedLoss(q,t)=(alpha+beta)/q*exp(−

qt)+alpha(t−

1/q), wherein expected loss is computed along a z-axis as a function of q (y-axis, from 0 to about

     2) and t (x-axis from 0 to about

     10).

30. A system to facilitate communications and collaboration between entities, comprising:

means for learning activity patterns from one or more user events;

means for generating predictions relating to future user states; and

means for predicting the future user states based upon an application a user currently is working on and the events, the future user states modeled in accordance with a continuous time Bayesian network.

31. A prediction system to facilitate operating systems management, comprising:

a processing component to determine a state of a computer; and

a model associated with the processing component that generates predictions relating to resource requirements of the computer, the model adapted in accordance with a continuous time Bayesian network.

32. The system of claim 31, the model determines allocation of resources in an operating system running on the computer, based on expectations under uncertainty on when different resources and information are needed by at least one component.

33. The system of claim 31, the model is employed to move components in memory, delete data from memory, close unused applications, adjust operating system performance, perform automated maintenance, perform cache management, adjust pre-fetching operations, adjust interrupt processing, initiate dynamic performance optimization or perform speculative computing operations.