Adaptive TV program recommender

• US 7,051,352 B1
• Filed: 02/04/2000
• Issued: 05/23/2006
• Est. Priority Date: 02/04/2000
• Status: Expired due to Fees

• Current Assignee: Sisvel S.p.A. (Fineur International S.A.)
• Original Assignee: Koninklijke Philips Electronics N.V. (Koninklijke Philips N.V.)
• Inventors: Schaffer, J. David
• Primary Examiner(s): Grant, Chris
• Assistant Examiner(s): HUYNH, AN SON PHI
• Application Number:
  US09/498,271
• Time in Patent Office: 2,300 Days
• Field of Search: 725/38, 725 9- 14, 725 44- 47, 706/45, 705/10, 705/14
• US Class Current: 725/39
• CPC Class Codes: H04N 21/4532 involving end-user characteristics, e.g. viewer profile, preferences monitoring of user activities for profile generation for accessing a video database G06F16/739; user profiles in network data switching protocols H04L67/306; processing of user preferences or user profiles in wireless networks H04W8/18
• CPC Class Codes: H04N 21/454 Content or additional data filtering, e.g. blocking advertisements
• CPC Class Codes: H04N 21/466 Learning process for intelligent management, e.g. learning user preferences for recommending movies services using the results of monitoring in broadcast systems H04H60/61
• CPC Class Codes: H04N 21/4663 involving probabilistic networks, e.g. Bayesian networks
• CPC Class Codes: H04N 21/4667 Processing of monitored end-user data, e.g. trend analysis based on the log file of viewer selections
• CPC Class Codes: H04N 21/4668 for recommending content, e.g. movies
• CPC Class Codes: H04N 21/4826 using recommendation lists, e.g. of programs or channels sorted out according to their score
• CPC Class Codes: H04N 7/163 by receiver means only

Abstract

A system for recommending television programs makes use of probabilistic calculations and a viewer profile to create a recommendation. The probabilistic calculations preferably are in the form of Bayesian classifier theory. Modifications to classical Bayesian classifier theory are proposed.

Claims

1. A data processing device comprising:

at least one input for receiving data including

viewer profile data; and

data regarding a television program;

a medium readable by the data processing device coupled to the input, said medium storing said viewer profile data; and

a processor, the processor being adapted to perform the following;

calculating a probability that the television program is a desired one; and

supplying a recommendation regarding the television program based on the probability,

wherein the processor maintains the viewer profile in accordance with a data structure comprising;

a list of feature values; and

for each element of the list, a respective number of times programs having that feature value were watched, and a respective number of times programs having that feature value were not watched,

and wherein the processor is further arranged to perform the following, each time a user watches a new program,

first adding, to the list, feature values or counts of such feature values, associated with that new program;

selecting at least one companion program to the new program, the companion program being selected at random from a program schedule, which companion program has not been watched; and

second adding, to the list, feature values of the companion program, or counts of such feature values.

2. The data processing device of claim 1, wherein the processor is further arranged to perform the following, each time a user watches a new program:

first adding, to the list, feature values or counts of such feature values, associated with that new program.

3. The data processing device of claim 1, wherein the input is a network connection.

4. The data processing device of claim 1, wherein calculating comprises using a Bayesian classifier.

5. The data processing device of claim 4, wherein the processor is further adapted to subject the viewer profile to a noise threshold calculation prior to using the Bayesian classifier.

6. A data processing device comprising:

at least one input for receiving data including

viewer profile data; and

data regarding a television program; and

a processor, the processor being adapted to perform the following;

calculating, using a Bayesian classifier, a probability that the television program is a desired one; and

supplying a recommendation regarding the television program based on the probability,

wherein the processor is further adapted to subject the viewer profile to a noise threshold calculation prior to using the Bayesian classifier,

and wherein

the viewer profile data comprises

a list of feature values;

a respective negative count for each element of the list, the negative count indicating a number of times programs having that feature value have not been watched;

a respective positive count for each element of the list, the positive count indicating a number of times programs having that feature value have been watched;

the noise threshold calculation comprises

selecting a sub-list comprising at least feature values having at least one specific type of feature;

choosing the highest negative count in the sub-list as the noise threshold;

the recommendation comprises a program selected from a group having at least one feature value having a positive or negative count in the viewer profile, which count exceeds the noise threshold.

7. The data processing device of claim 6, wherein the specific type comprises a day and time of day feature type.

8. The data processing device of claim 6, wherein the specific type comprises a station identification feature type.

9. The data processing device of claim 6, wherein the viewer profile data comprises a plurality of respective counts of programs watched, each respective count indicating how many programs watched had a respective feature.

10. The data processing device of claim 9, wherein calculating comprises calculating a probability that the television program is in a particular class.

11. The data processing device of claim 10, wherein the class is one of

programs the viewer is interested in, and

programs the viewer is not interested in.

12. A data processing device comprising:

at least one input for receiving data including

viewer profile data; and

data regarding a television program; and

a processor, the processor being adapted to perform the following;

calculating, using a Bayesian classifier, a probability that the television program is a desired one; and

supplying a recommendation regarding the television program based on the probability,

wherein the processor is further adapted to subject the viewer profile to a noise threshold calculation prior to using the Bayesian classifier,

and wherein subjecting the viewer profile to the noise threshold further comprises using observations gathered by a known random process to estimate a reasonable noise threshold.

13. A data processing device comprising:

at least one input for receiving data including

viewer profile data; and

data regarding a television program; and

a processor, the processor being adapted to perform the following;

calculating a probability that the television program is a desired one; and

supplying a recommendation regarding the television program based on the probability,

wherein calculating the probability comprises;

computing a prior possibility, of whether a program is desired or not;

computing a conditional probability of whether a feature fi will be present if a show is desired or not; and

computing a posterior probability of whether program is desired or not, based on the conditional probability and the prior probability.

14. The data processing device of claim 13, wherein it is assumed that programs watched are programs that the viewer is interested in.

15. The data processing device of claim 13, wherein the processor is further adapted to provide a recommendation regarding an additional item, other than a television program, based on the viewer profile.

16. A data processing device comprising:

at least one input for receiving data including

viewer profile data; and

data regarding a television program; and

a processor, the processor being adapted to perform the following;

calculating a probability that the television program is a desired one; and

supplying a recommendation regarding the television program based on the probability,

wherein

the viewer profile comprises a list of features types and values for such feature types;

the feature types are selected from at least two sets, including

a first set of feature types whose values are deemed non-independent; and

a second set of feature types whose values are deemed independent; and

calculating a probability comprises

applying a Bayesian classifier calculation corresponding to feature types from the second set; and

applying a modified Bayesian classifier calculation corresponding to feature types from the first set.

17. The data processing device of claim 16, wherein

with respect to features of the first set, the modified Bayesian classifier calculation considers only feature values that match with a show being classified.

18. A computer readable medium having computer-executable instructions stored thereon for performing the method comprising:

calculating a probability that a television program is a desired one, based on a viewer profile and data regarding the television program; and

supplying a recommendation regarding the television program based on the probability,

wherein the computer readable medium further embodies the viewer profile, the viewer profile being embodied as a data structure comprising;

a list of feature values; and

for each element of the list, a respective number of times programs having that feature value were watched,

and wherein the software is further arranged to perform the following, each time a user watches a new program,

first adding, to the list, feature values or counts of such feature values, associated with that new program;

selecting at least one companion program to the new program, the companion program being selected at random from a program schedule, which companion program has not been watched; and

second adding, to the list, feature values of the companion program, or counts of such feature values.

19. The computer readable medium of claim 18, wherein the computer-executable instructions is further arranged to perform the following, each time a user watches a new program:

first adding, to the list, feature values or counts of such feature values, associated with that new program.

20. The computer readable medium of claim 18, wherein the computer readable medium embodies the data regarding the television program.

21. The computer readable medium of claim 18, wherein calculating comprises using a Bayesian classifier.

22. The computer readable medium of claim 21, wherein the computer-executable instructions, is further adapted to subject the viewer profile to a noise threshold calculation prior to using the Bayesian classifier.

23. A computer readable medium having computer-executable instructions stored thereon for performing the method comprising:

calculating, using a Bayesian classifier, a probability that a television program is a desired one, based on a viewer profile and data regarding the television program; and

supplying a recommendation regarding the television program based on the probability,

wherein the computer-executable instructions is further adapted to subject the viewer profile to a noise threshold calculation prior to using the Bayesian classifier,

and wherein

the viewer profile data comprises

a list of feature values;

a respective negative count for each element of the list, the negative count indicating a number of times programs having that feature value have not been watched;

a respective positive count for each element of the list, the positive count indicating a number of times programs having that feature value have been watched;

the noise threshold calculation comprises

selecting a sub-list comprising at least feature values having at least one specific type of feature;

choosing the highest negative count in the sub-list as the noise threshold;

the recommendation comprises a program selected from a group having at least one feature value having a positive or negative count in the viewer profile exceeding the noise threshold.

24. The computer readable medium of claim 23, wherein the specific type comprises a day and time of day feature type.

25. The computer readable medium of claim 23, wherein the specific type comprises a station identification feature type.

26. The computer readable medium of claim 23, wherein the viewer profile data comprises a plurality of respective counts of programs watched, each respective count indicating how many programs watched had a respective feature.

27. The computer readable medium of claim 26, wherein calculating comprises calculating a probability that the television program is in a particular class.

28. The computer readable medium of claim 26, wherein the class comprises at least one of programs the viewer is interested in and programs the viewer is not interested in.

29. A computer readable medium having computer-executable instructions stored thereon for performing the method comprising:

calculating, using a Bayesian classifier, a probability that a television program is a desired one, based on a viewer profile and data regarding the television program; and

supplying a recommendation regarding the television program based on the probability,

wherein the computer-executable instructions is further adapted to subject the viewer profile to a noise threshold calculation prior to using the Bayesian classifier,

and wherein subjecting the viewer profile to the noise threshold further comprises using observations gathered by a known random process to estimate a reasonable noise threshold.

30. A computer readable medium having computer-executable instructions stored thereon for performing the method comprising:

calculating a probability that a television program is a desired one, based on a viewer profile and data regarding the television program; and

supplying a recommendation regarding the television program based on the probability,

wherein calculating the probability comprises;

computing a prior possibility, of whether a program is desired or not;

computing a conditional probability of whether a feature fi will be present if a show is desired; and

computing a posterior probability of whether program is desired or not, based on the conditional probability and the prior probability.

31. The computer readable medium of claim 30, wherein it is assumed that programs watched are programs that the viewer is interested in.

32. The computer readable medium of claim 30, wherein the computer-executable instructions is further arranged to provide a recommendation regarding an additional item, other than a television program, based on the viewer profile.

33. A computer readable medium having computer-executable instructions stored thereon for performing the method comprising:

calculating a probability that a television program is a desired one, based on a viewer profile and data regarding the television program; and

supplying a recommendation regarding the television program based on the probability,

wherein

the viewer profile comprises a list of features types and values for such feature types;

the feature types are selected from at least two sets, including

a first set of feature types whose values are deemed non-independent; and

a second set of feature types whose values are deemed independent; and

calculating a probability comprises

applying a Bayesian classifier calculation corresponding to feature types from the second set; and

applying a modified Bayesian classifier calculation corresponding to feature types from the first set.

34. The computer readable medium of claim 33, wherein with respect to features of the first set, the modified Bayesian classifier calculation considers only values that match with a show being classified.

35. A data processing method comprising performing the following operations in a data processing device:

first receiving data reflecting physical observations, which data includes a list of feature values and observations about feature values, some of which feature values are independent and some of which are not;

second receiving data about an item to be classified, the data about the item to be classified including feature values;

maintaining a division of the data reflecting physical observations into at least two sets, including

a first set including those feature values which are deemed not independent; and

a second set including those feature values which are deemed independent;

performing a probabilistic calculation on the data reflecting physical observations and the data regarding an item to be classified including;

applying a Bayesian classifier calculation with respect to feature values relating to the second set; and

applying a modified Bayesian classifier calculation with respect to feature values relating to the first set

presenting a conclusion regarding the item to be classified to a user based on the probabilistic calculation.

36. The method of claim 35, wherein the modified Bayesian classifier calculation comprises ignoring feature values from the data reflecting physical observations when those feature values are not present in the data regarding the item to be classified.