Method, system, and computer program product for adaptively learning user preferences for smart services

• US 8,041,658 B2
• Filed: 09/14/2009
• Issued: 10/18/2011
• Est. Priority Date: 12/31/2006
• Status: Expired due to Fees

• Current Assignee: Scenera Mobile Technologies LLC (Scenera Research LLC)
• Original Assignee: Ektimisi Semiotics Holdings, LLC
• Inventors: Singh, Munindar P., Singh, Mona
• Primary Examiner(s): Holmes, Michael B
• Application Number:
  US12/558,909
• Publication Number:
  US 20100070444A1
• Time in Patent Office: 764 Days
• Field of Search: 706/14
• US Class Current: 706/14
• CPC Class Codes: G06Q 30/02 Marketing; Price estimation or determination; Fundraising
• CPC Class Codes: H04L 41/12 Discovery or management of network topologies
• CPC Class Codes: H04L 41/147 for predicting network behaviour
• CPC Class Codes: H04L 41/16 using machine learning or artificial intelligence
• CPC Class Codes: H04L 41/5048 Automatic or semi-automatic definitions, e.g. definition templates
• CPC Class Codes: H04L 41/5054 Automatic deployment of services triggered by the service manager, e.g. service implementation by automatic configuration of network components
• CPC Class Codes: H04L 41/5067 Customer-centric QoS measurements
• CPC Class Codes: H04L 67/306 User profiles

Abstract

A method, system, and computer program product are described for adaptively learning user preferences for smart services. According to an exemplary embodiment, a method for adaptively learning user preferences for smart services includes modeling an availability of a subscriber for responding to an event associated with a service in terms subscriber context information available to determine a current situation of the subscriber related to the service, the subscriber context information based on private information of the subscriber. The availability of the subscriber for responding to the event is determined using a probability value associated with at least a portion of the subscriber context information. The probability value associated with the portion of the subscriber context information is updated based on feedback received from the subscriber in response to being presented a response to the event.

Claims

1. A method for adaptively learning user preferences for smart services, comprising:

modeling an availability of a subscriber for responding to an event associated with a service based on subscriber context information to determine a current situation of the subscriber, the subscriber context information based on private information of the subscriber;

determining the availability of the subscriber for responding to the event based on a probability value associated with at least a portion of the subscriber context information; and

updating the probability value associated with the at least a portion of the subscriber context information based on feedback received from the subscriber in response to being presented a response to the event.

2. The method of claim 1, wherein:

modeling the availability of the subscriber for responding to the event includes modeling the availability of a subscriber for responding to the event associated with the service based on probability values associated with attributes of the event and the subscriber context information; and

determining the availability of the subscriber for responding to the event includes determining the availability of the subscriber for responding to the event based on a probability value associated with an event attribute and the probability value associated with the at least a portion of the subscriber context information.

3. The method of claim 2, wherein modeling the availability of the subscriber for responding to the event comprises assigning a single-attribute probability distribution to the event attributes and portions of the subscriber context information that are unrelated in the model to other event attributes or other portions of the subscriber context information and assigning dual-attribute probability distributions to pairs among the event attributes and portions of subscriber context information related to one another in the model.

4. The method of claim 3, wherein the single-attribute probability distribution includes at least one of:

a point distribution including a one-dimensional array of probabilities having one element for each possible probability value of an associated event attribute or portion of the subscriber context information, wherein successive probability values stored in the array are unrelated to one another;

an n-gram distribution including a one-dimensional array of probabilities having one element for each possible probability value of an associated event attribute or portion of the subscriber context information, wherein successive probability values stored in the array are related to one another; and

a time distribution including a one-dimensional array of probabilities having one element for each possible probability value of an associated event attribute or portion of the subscriber context information, wherein successive probability values stored in the array are related to one another based on time.

5. The method of claim 4, wherein the dual-attribute probability distribution includes:

a pairwise distribution, based on at least one of the point, n-gram, and time distributions, the pairwise distribution including a two-dimensional array of probabilities having one element for each possible probability value of an associated pair among the event attributes and portions of the subscriber context information related to one another in the model.

6. The method of claim 3, comprising determining a probability value corresponding to a current value of an event attribute or a portion of the subscriber context information using the n-gram probability distribution by:

defining a weighting distribution including a one-dimensional array of weights having a width, midpoint, decay factor, and center weight, wherein the center weight is assigned to the midpoint of the weighting distribution and each of the remaining weights of the distribution is equal to the center weight divided by the product of the decay factor and a separation of the particular weight from the midpoint of the distribution; and

pairing corresponding values of the n-gram and weighting distributions by aligning the midpoint of the weighting distribution with the current value of the event attribute or portion of the subscriber context information; and

summing the products of the paired values of the n-gram and weighting distributions and dividing the resulting sum by the sum of the weights of included in the weighting distribution.

7. The method of claim 3, wherein determining the availability of the subscriber for responding to the event comprises:

for each event attribute and portion of the subscriber context information having an assigned single-attribute distribution,

determining the probability value assigned by the single-attribute distribution corresponding to a current value of the event attribute or portion of the subscriber context information;

multiplying the probability value assigned by the single-attribute distribution by a first variable representing odds of the subscriber being available for responding to the event, and

multiplying one minus the probability value assigned by the single-attribute distribution by a second variable representing odds of the subscriber being unavailable for responding to the event;

for each pair among the event attributes and portions of the subscriber context information related to one another in the model having an assigned dual-attribute distribution,

determining the probability value assigned by the dual-attribute distribution corresponding to a current pair of values of the paired event attributes and/or portions of the subscriber context information;

multiplying the probability value assigned by the dual-attribute distribution by the first variable representing the odds of the subscriber being available for responding to the event, and

multiplying one minus the probability value assigned by the dual-attribute distribution by the second variable representing the odds of the subscriber being unavailable for responding to the event; and

converting the odds represented by the first and second variables into a probability that the subscriber is available for responding to the event by dividing a final value of the first variable by a sum of final values of the first and second variables.

8. The method of claim 7, wherein the first and second variables are assigned values that are real numbers ranging from zero to one, each being assigned an initial value of one.

9. The method of claim 7, wherein updating at least one of the probability values comprises:

multiplying the resulting probability that the subscriber is available for responding to the event by a third variable representing either a rate of learning from positive feedback or a rate of learning from negative feedback from the subscriber;

adding the product of the resulting probability and the third variable to a fourth variable representing circumstances where the subscriber feedback indicates the subscriber is either available or unavailable for responding to the event; and

assigning the quotient of the sum of the product and the fourth variable and a sum of the third variable and a constant to each of the event attributes, portions of the subscriber context information, and pairs thereof used in determining the resulting probability that the subscriber is available for responding to the event.

10. The method of claim 9, wherein the third variable is assigned a default value of one-half, the fourth variable is assigned a value that is a real number ranging from zero to one, the fourth variable having a default value of one when representing circumstances where the subscriber feedback indicates the subscriber is available and a default value of zero when representing circumstances where the subscriber feedback indicates the subscriber is unavailable, and the constant is equal to one.

11. The method of claim 1, comprising allowing any probability values updated in the model to return to corresponding default values after at least one of a pre-determined amount of time and an occurrence of a pre-determined number of events.

12. A system for adaptively learning user preferences for smart services, comprising:

a user model component configured to model an availability of a subscriber for responding to an event associated with a service based on subscriber context information available to determine a current situation of the subscriber, the subscriber context information based on private information of the subscriber; and

a user agent component, operatively coupled to the user model component,

the user agent component configured to

determine the availability of the subscriber for responding to the event based on a probability value associated with at least a portion of the subscriber context information, and

update the probability value associated with the at least a portion of the subscriber context information based on feedback received from the subscriber in response to being presented a response to the event.

13. The system of claim 12, wherein:

the user model component is configured to model the availability of a subscriber for responding to the event associated with the service based on probability values associated with attributes of the event and the subscriber context information; and

the user agent component is configured to determine the availability of the subscriber for responding to the event based on a probability value associated with an event attribute and the probability value associated with the at least a portion of the subscriber context information.

14. The system of claim 13, wherein the user model component includes a single-attribute probability distribution assigned to the event attributes and portions of the subscriber context information that are unrelated in the user model component to other event attributes or other portions of the subscriber context information and a dual-attribute probability distribution assigned to pairs among the event attributes and portions of subscriber context information related to one another in the user model component.

15. The system of claim 14, wherein the single-attribute probability distribution includes at least one of:

a point distribution including a one-dimensional array of probabilities having one element for each possible probability value of an associated event attribute or portion of the subscriber context information, wherein successive probability values stored in the array are unrelated to one another;

an n-gram distribution including a one-dimensional array of probabilities having one element for each possible probability value of an associated event attribute or portion of the subscriber context information, wherein successive probability values stored in the array are related to one another; and

a time distribution including a one-dimensional array of probabilities having one element for each possible probability value of an associated event attribute or portion of the subscriber context information, wherein successive probability values stored in the array are related to one another based on time.

16. The system of claim 15, wherein the dual-attribute probability distribution includes:

a pairwise distribution, based on at least one of the point, n-gram, and time distributions, the pairwise distribution including a two-dimensional array of probabilities having one element for each possible probability value of an associated pair among the event attributes and portions of the subscriber context information related to one another in the user model component.

17. The system of claim 14, wherein the user agent component is configured to determine a probability value corresponding to a current value of an event attribute or a portion of the subscriber context information using the n-gram probability distribution by:

defining a weighting distribution including a one-dimensional array of weights having a width, midpoint, decay factor, and center weight, wherein the center weight is assigned to the midpoint of the weighting distribution and each of the remaining weights of the distribution is equal to the center weight divided by the product of the decay factor and a separation of the particular weight from the midpoint of the distribution; and

pairing corresponding values of the n-gram and weighting distributions by aligning the midpoint of the weighting distribution with the current value of the event attribute or portion of the subscriber context information; and

summing the products of the paired values of the n-gram and weighting distributions and dividing the resulting sum by the sum of the weights of included in the weighting distribution.

18. The system of claim 14, wherein in determining the availability of the subscriber for responding to the event, the user agent component is configured to:

for each event attribute and portion of the subscriber context information having an assigned single-attribute distribution,

determine the probability value assigned by the single-attribute distribution corresponding to a current value of the event attribute or portion of the subscriber context information;

multiply the probability value assigned by the single-attribute distribution by a first variable representing odds of the subscriber being available for responding to the event, and

multiply one minus the probability value assigned by the single-attribute distribution by a second variable representing odds of the subscriber being unavailable for responding to the event;

for each pair among the event attributes and portions of the subscriber context information related to one another in the user model component having an assigned dual-attribute distribution,

determine the probability value assigned by the dual-attribute distribution corresponding to a current pair of values of the paired event attributes and/or portions of the subscriber context information;

multiply the probability value assigned by the dual-attribute distribution by the first variable representing the odds of the subscriber being available for responding to the event, and

multiply one minus the probability value assigned by the dual-attribute distribution by the second variable representing the odds of the subscriber being unavailable for responding to the event; and

convert the odds represented by the first and second variables into a probability that the subscriber is available for responding to the event by dividing a final value of the first variable by a sum of final values of the first and second variables.

19. The system of claim 18, wherein the user agent component is configured to assign values to the first and second variables that are real numbers ranging from zero to one, each being assigned an initial value of one.

20. The system of claim 18, wherein in updating at least one of the probability values, the user agent component is configured to:

multiply the resulting probability that the subscriber is available for responding to the event by a third variable representing either a rate of learning from positive feedback or a rate of learning from negative feedback from the subscriber;

add the product of the resulting probability and the third variable to a fourth variable representing circumstances where the subscriber feedback indicates the subscriber is either available or unavailable for responding to the event; and

assign the quotient of the sum of the product and the fourth variable and a sum of the third variable and a constant to each of the event attributes, portions of the subscriber context information, and pairs thereof used in determining the resulting probability that the subscriber is available for responding to the event.

21. The system of claim 20, wherein the user agent component is configured to assign the third variable a default value of one-half, the fourth variable a real number ranging from zero to one having a default value of one when representing circumstances where the subscriber feedback indicates the subscriber is available and a default value of zero when representing circumstances where the subscriber feedback indicates the subscriber is unavailable, and the constant a value of one.

22. The system of claim 13, wherein the user agent component is configured to return any probability values updated in the user model component to corresponding default values after at least one of a pre-determined amount of time and an occurrence of a pre-determined number of events.

23. A computer readable medium containing a computer program, executable by a machine, for adaptively learning user preferences for smart services, the computer program comprising executable instructions for:

modeling an availability of a subscriber for responding to an event associated with a service based on subscriber context information available to determine a current situation of the subscriber, the subscriber context information based on private information of the subscriber;

determining the availability of the subscriber for responding to the event based on a probability value associated with at least a portion of the subscriber context information; and

updating the probability value associated with the at least a portion of the subscriber context information based on feedback received from the subscriber in response to being presented a response to the event.

24. A system for adaptively learning user preferences for smart services, comprising:

means for modeling an availability of a subscriber for responding to an event associated with a service based on subscriber context information available to determine a current situation of the subscriber, the subscriber context information based on private information of the subscriber;

means for determining the availability of the subscriber for responding to the event based on a probability value associated with at least a portion of the subscriber context information; and

means for updating the probability value associated with the at least a portion of the subscriber context information based on feedback received from the subscriber in response to being presented a response to the event.