Viewing behavior learning apparatus

A method for learning a viewing behavior for a program including at least one of video or audio, comprising:

• electronically acquiring, from a television, a viewing record indicating an attribute of the program and a viewing time point indicating a start time and an end time during which the program was viewed;
• dividing the viewing record into divided viewing records according to discretized time points discretized by a unit time, wherein the divided viewing records respectively denote a viewing time period during which the program was viewed;
• discretizing the viewing time period for the respective divided viewing records by an interval shorter than the unit time to yield discretized viewing time periods;
• replicating the divided viewing records to yield a number of replicated viewing records equal to a number of the discretized viewing time periods;
• determining respective viewing frequency values for the divided viewing records based on the replicated viewing records;
• building a model of the viewing behavior by use of a Bayesian network having the viewing time point and the attribute of the program as random variables;
• storing a conditional probability table of the Bayesian network;
• and
• updating a subset of records of the conditional probability table corresponding to the divided viewing records using the respective viewing frequency values.

A non-transitory computer-readable memory having stored thereon computer-executable instructions that, in response to execution, cause a computer system to perform operations, comprising:

• acquiring a viewing record indicating an attribute of a program and a viewing time point indicating a start time and an end time during which the program was viewed;
• dividing the viewing record into divided viewing records according to discretized time points discretized by a unit time, wherein the divided viewing records denote respective viewing time periods during which the program was viewed;
• discretizing the viewing time periods for the respective divided viewing records by an interval shorter than the unit time to yield discretized viewing time periods;
• replicating the divided viewing records to yield a number of replicated viewing records equal to a number of the discretized viewing time periods;
• determining respective viewing frequency values for the divided viewing records based on the replicated viewing records;
• building a model of the viewing behavior by use of a Bayesian network having the viewing time point and the attribute of the program as random variables;
• storing a conditional probability table of the Bayesian network;
• and
• updating a subset of records of the conditional probability table corresponding to the divided viewing records using the respective viewing frequency values.