KR20130041725A - System for recommending favorite channel/program based on tv watching pattern and method thereof - Google Patents

System for recommending favorite channel/program based on tv watching pattern and method thereof Download PDF

Info

Publication number
KR20130041725A
KR20130041725A KR1020120088364A KR20120088364A KR20130041725A KR 20130041725 A KR20130041725 A KR 20130041725A KR 1020120088364 A KR1020120088364 A KR 1020120088364A KR 20120088364 A KR20120088364 A KR 20120088364A KR 20130041725 A KR20130041725 A KR 20130041725A
Authority
KR
South Korea
Prior art keywords
program
channel
probability value
preference
candidate
Prior art date
Application number
KR1020120088364A
Other languages
Korean (ko)
Inventor
조정우
Original Assignee
한국전자통신연구원
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 한국전자통신연구원 filed Critical 한국전자통신연구원
Priority to US13/653,826 priority Critical patent/US8789109B2/en
Publication of KR20130041725A publication Critical patent/KR20130041725A/en

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
    • H04N21/20—Servers specifically adapted for the distribution of content, e.g. VOD servers; Operations thereof
    • H04N21/25—Management operations performed by the server for facilitating the content distribution or administrating data related to end-users or client devices, e.g. end-user or client device authentication, learning user preferences for recommending movies
    • H04N21/258—Client or end-user data management, e.g. managing client capabilities, user preferences or demographics, processing of multiple end-users preferences to derive collaborative data

Abstract

PURPOSE: A system for recommending preferred channels/programs by considering TV watching patterns and a method thereof are provided to recommend preferred channels/programs by extracting the preferred channels/programs through the application of an entropy theory to a Bayesian network, and applying fuzzy logic calculation to the preferred channels/programs. CONSTITUTION: A Bayesian network learning unit(110) learns a Bayesian network based on inputted watching pattern information. A candidate preferred channel/program extracting unit(120) extracts a candidate preferred channel/program based on a first probability value. A fuzzy logic calculation unit(130) performs fuzzy logic inference calculation for the extracted candidate preferred channel/program. A preferred channel/program recommending unit(140) determines a preferred channel/program based on the first probability value and a second probability value. [Reference numerals] (110) Bayesian network learning unit; (120) Candidate preferred channel/program extracting unit; (130) Fuzzy logic calculation unit; (140) Preferred channel/program extracting unit;

Description

System and method for recommending preferred channels / programs in consideration of TV viewing pattern {SYSTEM FOR RECOMMENDING FAVORITE CHANNEL / PROGRAM BASED ON TV WATCHING PATTERN AND METHOD THEREOF}

The present invention relates to a method for recommending a preferred channel / program, and learning a TV watching pattern using a Bayesian network, extracting candidate preferred recommendation channels / programs by applying entropy theory to the learned Bayesian network, and extracting candidates. The present invention relates to a system and a method for recommending a favorite channel / program by applying fuzzy logic operations to the favorite channel / program.

As TV channels and types of programs increase, users have more choices, but they have to find the TV channels and programs they want. As a solution to this hassle, a solution using Bayesian Network theory has been proposed. This method performs user TV viewing pattern learning in terms of probability and recommends user-specific TV preference channels / programs based on the learned probability values.

However, the existing Bayesian Network theory approach has the following problems.

The purpose of the TV is to provide leisure. However, for example, when watching EBS broadcasting in preparation for college entrance examination, when watching TV for leisure purposes, EBS channels with high viewing probability are highly recommended, unlike the user's intended intention. On the contrary, if you are too busy to watch your favorite TV shows for a while, you may not be able to get out of the rankings because of low viewing probability. As a result, an optimal recommendation that meets the user's intention is not made.

Accordingly, an object of the present invention is to apply a Bayesian network to learn TV viewing patterns, to apply entropy theory to the learned Bayesian network, and to extract and extract candidate preferred recommendation channels / programs. The present invention provides a system and method for recommending a preferred channel / program by applying a fuzzy logic operation to the candidate candidate channel / program.

However, the objects of the present invention are not limited to those mentioned above, and other objects not mentioned can be clearly understood by those skilled in the art from the following description.

In order to achieve the above objects, a system for recommending a preferred channel / program according to an aspect of the present invention learns a Bayesian network based on the viewing pattern information of an input user, and as a result of the learning, A Bayesian network learner for calculating a first probability value; A candidate preference channel / program extraction unit configured to extract a candidate preference channel or a candidate preference program based on the calculated first probability value; A fuzzy logic calculator configured to perform fuzzy logic inference on the extracted candidate preference channel or candidate preference program and calculate a second probability value for each candidate preference channel or candidate preference program as a result of the fuzzy logic inference operation; And a preference channel / program recommendation unit for determining a preference channel or a preference program according to the viewing pattern information of the user based on the calculated first probability value and the second probability value.

Preferably, the candidate selection channel / program extractor applies an entropy function to a first probability value calculated from the Bayesian network learner, and when the result of the application is less than or equal to a preset threshold, Some or all of a channel or program corresponding to a probability value may be extracted as the candidate preferred channel or candidate preferred program.

Preferably, the fuzzy logic operation unit performs a fuzzy logic operation based on preferences and preset rules received from a user for each candidate preference channel or candidate preference program, and as a result of performing the candidate preference channel or candidate preference. A second probability value for each program is calculated.

Preferably, the fuzzy logic calculator comprises a fuzzy equalizer for performing a fuzzy to give a degree of belonging to the preference input from the user and output a fuzzy value as a result of the fuzzy; A fuzzy inference unit for performing fuzzy inference based on the output fuzzy value and predetermined rules; And a fuzzy purifier for performing a fuzzy operation on the result value obtained as a result of the fuzzy inference and outputting a second probability value, which is a fuzzy value as a result of the fuzzy inference.

Preferably, the preference is characterized in that the information indicating the preference or unfavorable for each channel or program received from the user.

Preferably, the predetermined rules are characterized in that the IF-THEN rule that is a statement indicating a relationship between a series of facts.

Preferably, the preferred channel / program recommender uses a first probability value of a channel or program corresponding to the second probability value among all channels or programs as the second probability value, and proportionally calculates the rest based on the second probability value. The first probability value of the channel or program is updated to determine and recommend a preferred channel or program based on the updated first probability value.

Preferably, the viewing pattern information includes the amount of TV viewing time by day, the amount of TV viewing time by time zone, the total viewing time by channel, and the viewing time by program genre.

According to another aspect of the present invention, there is provided a method for recommending a preferred channel / program by learning a Bayesian network based on inputted viewing pattern information of a user and calculating first probability values for all channels or programs as a result of the learning. step; Extracting a candidate preference channel or a candidate preference program based on the calculated first probability value; Performing a fuzzy logic inference operation on the extracted candidate preference channel or candidate preference program and calculating a second probability value for each candidate preference channel or candidate preference program as a result of the fuzzy logic inference operation; And determining a preference channel or a preference program according to the viewing pattern information of the user based on the calculated first probability value and the second probability value.

Preferably, the extracting may include applying an entropy function to a first probability value calculated from the Bayesian network learner, and when the result of the entropy function is less than or equal to a preset threshold, the first probability value corresponds to the first probability value. Some or all of the channels or programs to be extracted as the candidate preference channel or candidate preference program.

Preferably, the calculating of the second probability value may perform a fuzzy logic operation based on a preference received from a user and preset rules for each of the candidate preference channel or candidate preference program, and as a result of performing the candidate preference. A second probability value for each channel or candidate preference program is calculated.

Preferably, the calculating of the second probability value comprises performing fuzzy to give a degree of belonging to the preference input from the user and outputting a fuzzy value as a result of the fuzzy; Performing fuzzy inference based on the output fuzzy values and predetermined rules; And performing a dispersing on the result value obtained as the result of the fuzzy inference, and outputting a second probability value that is unpurged as the result of the fuzzy inference.

Preferably, the preference is characterized in that the information indicating the preference or unfavorable for each channel or program received from the user.

Preferably, the predetermined rules are characterized in that the IF-THEN rule that is a statement indicating a relationship between a series of facts.

Preferably, the determining may include using a first probability value of a channel or program corresponding to the second probability value among all channels or programs as the second probability value, and proportionally calculating the second probability value based on the second probability value. The first probability value of the program is updated to determine and recommend a favorite channel or a program based on the updated first probability value.

Preferably, the viewing pattern information includes the amount of TV viewing time by day, the amount of TV viewing time by time zone, the total viewing time by channel, and the viewing time by program genre.

Through this, the present invention learns TV viewing pattern using Bayesian network, extracts the candidate preference recommendation channel / program by applying entropy theory to the learned Bayesian network, and performs fuzzy logic operation on the extracted candidate preference channel / program. By recommending a preferred channel / program by applying, it is possible to efficiently extract the preferred channel / program.

1 is a diagram illustrating a system for recommending a preferred channel / program according to an embodiment of the present invention.
2 is a diagram illustrating a structure of a Bayesian network according to an embodiment of the present invention.
3 is a diagram showing a graph of the entropy formula for the probability that the front and the back of the coin when the coin is thrown indefinitely according to an embodiment of the present invention.
4 illustrates a graph of a trigonometric function according to an embodiment of the present invention.
FIG. 5 is a diagram illustrating definitions of five language variables belonging to a recommendation item of a THEN backrest unit according to an exemplary embodiment of the present invention.
6 shows a detailed configuration of a fuzzy logic calculating unit according to an embodiment of the present invention.
7 illustrates a possible linear reduction function according to an embodiment of the present invention.
8 illustrates a method for recommending a preferred channel / program according to an embodiment of the present invention.

Hereinafter, a system and method for recommending a preferred channel / program in consideration of a TV viewing pattern according to an embodiment of the present invention will be described in detail with reference to FIGS. 1 to 8. Reference numerals used throughout the specification represent the same components as the reference numerals shown in each drawing. In addition, in describing the present invention, when it is determined that a detailed description of a related known function or configuration may unnecessarily obscure the subject matter of the present invention, a detailed description thereof will be omitted.

In particular, the present invention uses a Bayesian network to learn TV viewing patterns, apply entropy theory to the learned Bayesian network, extract candidate preferred recommendation channels / programs, and apply fuzzy logic operations to the extracted candidate preference channels / programs. We propose a new method for determining and recommending preferred channels / programs.

1 is a diagram illustrating a system for recommending a preferred channel / program according to an embodiment of the present invention.

As shown in FIG. 1, a system for selecting a preferred channel / program according to the present invention includes a Bayesian network learner 110, a candidate preferred channel / program extractor 120, a fuzzy logic operator 130, and a preference. Channel / program recommender 140 and the like.

In this case, the system for recommending a preferred channel / program according to the present invention may operate by collecting channel data input by a user.

The Bayesian network learner 110 may receive preset viewing pattern information, for example, the amount of TV viewing time by day, the amount of TV viewing time by day, the total viewing time by channel, and the viewing time by program genre. Such viewing pattern information may be increased or decreased as necessary.

The Bayesian network learning unit 110 may learn a previously designed Bayesian network using the received viewing pattern information, which will be described with reference to FIG. 2.

2 is a diagram illustrating a structure of a Bayesian network according to an embodiment of the present invention.

As illustrated in FIG. 2, the Bayesian network learning unit 110 may use, for example, a TV viewing time amount D for each day, a TV viewing time amount T for each time zone, a total viewing time amount C for each channel, and a viewing time amount G for each program genre. Bayesian networks can be configured.

For example, the day of the week is divided into Mondays, Tuesdays, Wednesdays, Thursdays, Fridays, Saturdays, and Sundays. It can be divided into 20 o'clock, 21 o'clock to 24 o'clock at night.

In this case, since the structure of the Bayesian network may be changed according to the input viewing pattern information, the Bayesian network is not limited to a specific form and comprehensively includes other forms of design.

 The Bayesian network structure constructed by the Bayesian network learning unit 110 may be defined as a probability equation considering all of the viewing pattern information D, T, C, and G as shown in Equation 1 below. That is, the Bayesian network learning unit 110 uses the input viewing pattern information to input four right probability terms p (D), p (T | D), p (C | D, T), Obtain p (G | D, T, C), multiply p (D), p (T | D), p (C | D, T), p (G | D, T, C) 1] will learn the left probability term.

[Equation 1]

p (D, T, G, C) = p (D) p (T | D) p (C | D, T) p (G | D, T, C)

Here, D denotes the amount of TV viewing time by day, T denotes the amount of TV viewing time by time zone, C denotes the total amount of viewing time by channel, and G denotes the amount of viewing time by program genre.

Accordingly, the Bayesian network learner 110 may calculate a first probability value p (D, T, G, C) for each channel or program as a result of the learning.

The candidate preference channel / program extractor 120 may extract the candidate preference channel or the candidate preference program based on the first probability value calculated from the Bayesian network learner 110.

In order to extract candidate preference channels / programs, the present invention utilizes the concept of entropy. This entropy function H (X) can be expressed as Equation 2 below.

&Quot; (2) "

Figure pat00001

Where k denotes a variable that the designer sets so that the maximum value of the entropy function is always maintained at 1, x denotes a random variable, n denotes the number of maximum random variables, and b denotes a variable set by the designer as needed. 2 is usually used as a variable.

3 is a diagram showing a graph of the entropy formula for the probability that the front and the back of the coin when the coin is thrown indefinitely according to an embodiment of the present invention.

As shown in FIG. 3, the closer the result H (X) of the entropy function is to 1, the more difficult it is to predict whether or not an arbitrary event will occur, and the closer to 0, the easier it is to predict whether an event will occur. . In other words, if the H (X) is less than or equal to the threshold as a result of applying the entropy function to some or all of the probability values obtained from the Bayesian network, it can be said that the tendency of the probability value is derived from the Bayesian network.

If the prediction is possible, the channel / program having a corresponding probability value may be a candidate favorite channel / program.

At this time, the threshold value can be determined and changed by the designer through experimentation.

In the coin tossing, the probability value is generated in front of the coin and the back of the coin, whereas the probability value in the Bayesian network is the number of channels and corresponding programs. There may be 5 to 6 programs per channel.

However, if the result of sub-threshold is obtained by applying the entropy function, the coin toss can be judged whether the coin tends to be more or less biased, but for TV channels and programs, it can be judged how many items are skewed. There is no standard that can be. In this case, if the result of entropy function falls below the threshold, the number of candidate preferred channels / programs should be determined through the following order, when there are three or more items that are assigned to the entropy function operation: The maximum value is greater than one. In this case, set the k value in the entropy function to be the inverse of the maximum value so that the maximum value is always 1, no matter how many items are assigned.

Then set the interval by dividing between 0 and 1 by the number of items you assign to the entropy function. For example, if there are three items to assign to the entropy function, that is, the number of channels or programs, three intervals: [0,1 / 3], [1 / 3,2 / 3], [2 / 3,1] Can be divided into

Here, the threshold value was set to 2/3, that is, 0.6667. This value can be changed by the designer as needed.

1) If the entropy function calculation result is greater than 2/3 (0.6667), it can be determined that there are no significant biases or tendencies in the probability values of the three channel or program items.

2) If the entropy function grinding result is less than 2/3 (0.6667), it can be determined that there is a significant bias or tendency in the probability values of three channel or program items. In this case, it is necessary to determine how many of the three channels or programs will be presented to the user as meaningful channels / programs.

If the result of the entropy function calculation is between the intervals larger than 1/3 (0.3333) and smaller than 2/3 (0.6667), the number of candidate preference channels or candidate preference program items is set to two. The two items are set in order of increasing probability.

On the other hand, if the entropy function killing result is greater than 0 and less than 1/3 (0.3333), the number of candidate preference channels or candidate preference program items is set to one.

The fuzzy logic calculator 130 may perform fuzzy logic inference on a candidate preferred channel or a candidate preferred program and calculate a second probability value for each candidate preferred channel or a candidate preferred program as a result of the fuzzy logic operation. In operation 130, the user inputs information indicating a preference or non-preferred preference that means a user's preference, for example, good or dislike, for the candidate preference channel or candidate preference program obtained as a result of the entropy function. Here, preference is often expressed in the form of human language, such as good, dislike, slightly good, slightly dislike, better, less good.

In the present invention, fuzzy logic that can effectively deal with the uncertainty of language can be applied to deal with preferences through such a person's language. Choose between 0 and 1 as close to 0 if you want to be unfavorable, or close to 1 if you want to indicate a preference. In some cases, 0 may be selected to indicate the least preferred and 1 to indicate the highest preference.

Rules are set to handle the preferences entered by the user. The rule consists of IF-THEN type, and the IF part is called the front part and the THEN part is the back part. Set the language variable as a condition in the front part and set the language variable corresponding to the result in the back part.

At this time, each language variable is composed of a membership function. The trigonometric function among the membership functions can be expressed as Equation 3 below.

&Quot; (3) "

Figure pat00002

4 is a diagram illustrating a graph of a trigonometric function according to an embodiment of the present invention.

As shown in FIG. 4, in the present invention, a trigonometric function based on Equation 3 is applied as a membership function, but not necessarily limited thereto, and a Gaussian function or a trapezoidal function may also be applied.

In addition, the membership function in the drawing may be a type-1 membership function in a form shown in a two-dimensional plane, but may include, but is not limited to, type-2 and interval type-2 membership functions.

Once the membership function is determined, the rules are set. The designed rules can be expressed as shown in [Table 1].

1.IF P is High AND M is High, THEN R is High 2.IF P is Medium AND M is High, THEN R is Low 3.IF P is Low AND M is High, THEN R is Very_Low 4.IF P is High AND M is Medium, THEN R is High 5.IF P is Medium AND M is Medium, THEN R is Medium 6.IF P is Low AND M is Medium, THEN R is Low 7.IF P is High AND M is Low, THEN R is Very_High 8.IF P is Medium AND M is Low, THEN R is High 9.IF P is Low AND M is Low, THEN R is Low

The rules consist of the Preference (P) and Mandatori (M) items in the IF Department, and each item is composed of three language variables: High, Medium, and Low. The THEN Ministry of Public Affairs consists of Recommendation (R) items and consists of five language variables, Very High, High, Medium, Low, Very Low. The IF front and THEN back and forth sections, and the language variable assignment by item can be changed by the designer as needed.

FIG. 5 is a diagram illustrating definitions of five language variables belonging to a recommendation item of a THEN backrest unit according to an exemplary embodiment of the present invention.

As shown in FIG. 5, five language variable membership functions, for example, Very High, High, Medium, Low, Very Low, etc. for the Recommendation item of the THEN back office are shown.

The fuzzy logic operation unit 130 performs fuzzy logic inference operation based on the user input for the candidate preferred channel or the candidate preferred program and preset rules, and as a result of performing the fuzzy logic inference for each candidate preferred channel or candidate preferred program, The second probability value may be calculated.

The reason for the reasoning operation is that there are more cases where the language variable does not match between the preference entered by the user and the rules set by the designer. If the language variable set by the user and the language variable in the rule set by the designer are different, the value between them can be inferred through fuzzy logic inference.

6 shows a detailed configuration of a fuzzy logic calculating unit according to an embodiment of the present invention.

As shown in FIG. 6, the fuzzy logic calculating unit 130 according to the present invention may include a fuzzy purifier 131, a fuzzy reasoning unit 132, a non-fuzzy purifier 133, and the like.

The purifier 131 may receive a preference from a user with respect to the candidate preference channel or the candidate preference program. The value input here, that is, the preference may have a real form.

At this time, since the input preference is a form that cannot be handled by the fuzzy logic, it should be transformed into a form that can be handled by the fuzzy logic. This process is called fuzzy.

Accordingly, the fuzzy purifier 131 may perform fuzzy to give a membership grade to the input preference, and output a fuzzy value as a result of the purge.

The fuzzy inference unit 132 may perform fuzzy inference based on the output fuzzy values and rules. That is, the fuzzy reasoning unit 132 uses the result of the THEN back building unit as it is when the language variables in the IF front section match the fuzzy values input to the system in the rule.

In most cases, the fuzzy values do not exactly match the language variables set in the IF key section of the rule.If the language variables in the IF key section of the rule do not match the fuzzy values entered into the system, You cannot use the result as it is. In this case, if a fuzzy value that is slightly different from the preset language variable is input from the IF system, it may be very useful to infer a slightly different result based on the set rules. This process is performed by the fuzzy inference unit 132. That is, the fuzzy inference unit 132 performs the fuzzy inference when the rule “IF A, THEN B” is set by the designer and A 'is input as the fuzzy value, and B' is used as the result value. You can output

The process of performing a specific fuzzy inference can be expressed as Equation 4 below.

&Quot; (4) "

Figure pat00003

Where sup [] means to find the largest value among the membership degrees of the membership function, and ★ is a T-Norm operation that takes a smaller value by comparing the membership degrees of each value of the x coordinate with respect to two membership functions. T denotes the case where there are a plurality of T-Norm operations,-> denotes the operation between the IF front part and THEN back part in one rule, and sup is the largest value among the degree of belonging of the membership function. Meaning, B represents the full range of the result of the i th rule, F represents the full range of values that the IF front part can have, G represents the full range of values that the THEN back part can have, p represents the total number of repeated T-Norm operations.

In this case, since the result value obtained from the fuzzy inference operation, that is, the membership function is a value that cannot directly reflect the preference, one probability value that can be used in the existing number system must be calculated from the membership function. This process is called defuzzy.

Accordingly, the non-fuzzy diffuser 133 may perform the non-fuzzy operation on the result value output from the fuzzy inference unit 132 and output the second probability value that is unfuzzy as the result.

In the present invention, among the various methods for such dispersing, the second probability value y (x) obtained by using the center of gravity method and the dispersing process can be expressed by Equation 5 below.

[Equation 5]

Figure pat00004

Here, y (x) represents a function from which a result value is derived, and x represents an input value input to the result function.

At this time, in the present invention, the center of gravity method is applied among the methods for non-fuzzy, but not necessarily limited thereto, other methods for non-fuzzy can be applied as necessary.

The preference channel / program recommendation unit 140 based on the first probability value calculated from the Bayesian network learning unit 110 and the second probability value calculated from the fuzzy logic operation unit 130, according to a user's viewing pattern or preference program according to a viewing pattern of the user. Can finally be determined and recommended.

In this case, an embodiment of a combining algorithm between the specific first probability value and the second probability value is as follows.

When the second probability value is obtained from the fuzzy logic inference unit 140 reflecting the clear preferences of the user, the final value of the Bayesian network learner 110 is reflected after applying the value to the step before the final probability value calculation is performed in the Bayesian network learner 110. Calculate the probability value. Based on the calculated result, the new recommendation channel / program is ranked to provide a recommendation to the user.

In this case, it should be noted that when applying the second probability value to the Bayesian network learning unit 110, if the second probability value that is different from the first probability value obtained through the existing Bayesian network learning is used, the total probability value is 1. It can happen if In this case, since the system of the whole system is destroyed, when the second probability value is applied to the Bayesian network learning unit 110, an appropriate algorithm capable of maintaining the system of the whole system is required.

In an embodiment of the present invention, a proportional operation algorithm to be described below is used. For example, it is assumed that a first probability value obtained through the Bayesian network learner 110 is as shown in Table 2 below for a channel or program item composed of three probability values.

First probability value 0.1 (10%) 1.00 (100%) 0.3 (30%) 0.6 (60%)

It is assumed that the item selected by applying the entropy function is 0.3 (30%) item among them.

For example, a user inputs a preference for a meaningful channel / program item extracted through an entropy function, performs predefined rules and fuzzy logic inference calculations, and then uses a fuzzy operation to finally obtain a second probability value. For example, assume 0.72 (72%).

As shown in the following [Table 3], if the second probability value is replaced with 0.3 (30%), the total probability value exceeds 1 (100%), which may cause a problem in the system.

0.1 (10%) 1.42 (142%) 0.72 (72%) 0.6 (60%)

Since the second probability value is a certain preference entered directly by the user, this value cannot be sacrificed. On the other hand, since the remaining probability values are determined when there is an existing first probability value whose meaning is weak, these values should be changed so that the whole system does not exceed one.

The specific algorithm is as follows. Subtract 1 from the second probability of 0.72. The probability value obtained by the operation is multiplied by the ratio of the remaining first probability value to newly set the probability value as shown in Equation 6 below.

&Quot; (6) "

Figure pat00005

Computing according to the above algorithm maintains the total probability value 1 while reflecting the second probability value as shown in Table 4 below.

0.04 (4%) 1.00 (100%) 0.72 (72%) 0.24 (24%)

The system reflecting the second probability value is then updated in real time with the new probability value again through Bayesian network learning. In this case, the newly learned Bayesian network probability value may invalidate a probability value based on a user's previously entered preference. In order to prevent this phenomenon, if the second probability value is fixed invariably, a problem arises in that it cannot flexibly reflect the user's preference change.

In order to solve this problem, an algorithm corresponding to one embodiment is provided.

Sets a linear reduction function whose x and y axes are in the range of 0 to 1, respectively. Any type of function can be designed.

7 illustrates a possible linear reduction function according to an embodiment of the present invention.

As shown in Fig. 7, the partitions are divided at a constant ratio between 0 and 1 on the x axis. For example, it can be divided into 10 sections by 0.1. The number of these compartments can be flexibly adjusted by the designer as needed.

Each time the whole process of the system is repeated, the value of the x-axis is set starting from the first interval value after 0 and the value of y is obtained. For example, when the entire second process is repeated after the first second probability value is reflected, if the x-axis is divided into 10 sections by 0.1, the y value at that time becomes 0.9. After that, if the second whole operation is repeated, x is 0.2 and y is 0.8.

The new second probability value is set by multiplying the obtained y value by the existing second probability value. In this case, the second probability value to which the user directly applies the preference gradually decreases over time. If you want to maintain the influence of the second probability value over a long time, you can divide the x-axis section into small pieces.

The new second probability value and the first probability value updated in real time through Bayesian network learning are compared with each other.

If the new second probability value is larger than the updated first probability value, the new new second probability value is used instead of the real time updated first probability value. Since the new second probability value is different from the existing second probability value, the neighboring probability values are also changed to match the new second probability value. This keeps the overall probability value at 1. The applied algorithm may use the above-mentioned proportional operation algorithm. In this case, the probability values used for the proportional operation are based on the first probability values updated in real time.

On the other hand, when the new second probability value is less than or equal to the first probability value updated in real time, the new second probability value is discarded and the first probability value updated in real time through Bayesian network learning is used.

8 illustrates a method for recommending a preferred channel / program according to an embodiment of the present invention.

As shown in FIG. 8, the system for recommending a preferred channel / program according to the present invention (hereinafter referred to as a recommendation system) includes viewing pattern information, for example, the amount of TV viewing time D for each day of the week, and the amount of TV viewing time for each time of day. When receiving T, the total viewing time amount C for each channel, and the viewing time amount G for each program genre, the Bayesian network configured by the designer may be trained using the received viewing pattern information (S810).

Next, the recommendation system may calculate a first probability value for each of all channels or programs as a result of the learning (S820).

Next, the recommendation system may extract the candidate preference channel or the candidate preference program based on the calculated first probability value (S830).

Next, the recommendation system may present the extracted candidate preference channel or candidate preference program to the user, and may receive, for example, preference of each candidate preference channel or candidate preference program from the user (S840).

Next, the recommendation system performs a fuzzy logic operation on the candidate preference channel or the candidate preference program using the preferences input from the user and the preset rules, and as a result of this, the second recommendation system for each candidate preference channel or candidate preference program is performed. The probability value may be calculated (S850).

Next, the recommendation system may finally determine and recommend the preferred channel or the preferred program according to the viewing pattern information of the user based on the calculated first probability value and the second probability value (S860).

Meanwhile, the above-described embodiments of the present invention can be written as a program that can be executed in a computer, and can be implemented in a general-purpose digital computer that operates the program using a computer-readable recording medium. The computer-readable recording medium includes a storage medium such as a magnetic storage medium (e.g., ROM, floppy disk, hard disk, etc.), optical reading medium (e.g., CD ROM,

It will be apparent to those skilled in the art that various modifications and variations can be made in the present invention without departing from the spirit or essential characteristics thereof. Therefore, the embodiments disclosed in the present invention are intended to illustrate rather than limit the scope of the present invention, and the scope of the technical idea of the present invention is not limited by these embodiments. The protection scope of the present invention should be interpreted by the following claims, and all technical ideas within the equivalent scope should be interpreted as being included in the scope of the present invention.

110: Bayesian Network Learning Division
120: candidate preference channel / program extracting unit
130: fuzzy logic operation unit
131: purifier
132: fuzzy reasoning
133: befuger
140: preferred channel / program recommendation unit

Claims (16)

A Bayesian network learning unit learning a Bayesian network based on the received viewing pattern information of the user and calculating first probability values for all channels or programs as a result of the learning;
A candidate preference channel / program extraction unit configured to extract a candidate preference channel or a candidate preference program based on the calculated first probability value;
A fuzzy logic calculator configured to perform fuzzy logic inference on the extracted candidate preference channel or candidate preference program and calculate a second probability value for each candidate preference channel or candidate preference program as a result of the fuzzy logic inference operation; And
A preference channel / program recommender for determining a preference channel or a preference program according to the viewing pattern information of the user based on the calculated first probability value and the second probability value;
System for recommending a preferred channel / program comprising a. The method according to claim 1,
The candidate selection channel / program extraction unit,
Apply an entropy function to a first probability value calculated from the Bayesian network learner,
If the result of the entropy function is less than or equal to a predetermined threshold as a result of the application, part or all of the channel or program corresponding to the first probability value is extracted as the candidate preferred channel or candidate preferred program. System for recommending a program. The method according to claim 1,
The fuzzy logic operation unit,
A fuzzy logic operation is performed on each of the candidate preference channels or candidate preference programs based on preferences and preset rules received from a user;
And a second probability value for each candidate preference channel or candidate preference program as a result of the execution. The method of claim 3,
The fuzzy logic operation unit,
A fuzzy purifier for performing a fuzzy to give a degree of belonging to the preference input from the user and outputting a fuzzy value as a result of performing the fuzzy;
A fuzzy inference unit for performing fuzzy inference based on the output fuzzy value and predetermined rules; And
A non-fugeizer for performing a fuzzy operation on the result value obtained as a result of the fuzzy inference and outputting a second probability value that is unpurged as a result of the fuzzy inference;
System for recommending a preferred channel / program comprising a. The method of claim 3,
The preference is,
System for recommending a preferred channel / program, characterized in that the information indicating the preferences or preferences for each channel or program received from the user. The method of claim 3,
The preset rules refer to an IF-THEN rule, which is a sentence indicating a relationship between a series of facts. The method according to claim 1,
The preferred channel / program recommendation unit,
Using the first probability value of the channel or program corresponding to the second probability value among all channels or programs as the second probability value,
Recommend a preferred channel / program characterized in that the proportional calculation based on the second probability value to update the first probability value of the remaining channel or program to determine and recommend the preferred channel or program based on the updated first probability value System for doing so. The method according to claim 1,
The viewing pattern information,
A system for recommending a preferred channel / program, comprising the amount of TV viewing time by day, the amount of TV viewing time by hour, the total viewing time by channel, and the viewing time by program genre. Learning a Bayesian network based on the received viewing pattern information of the user and calculating first probability values for all channels or programs as a result of the learning;
Extracting a candidate preference channel or a candidate preference program based on the calculated first probability value;
Performing a fuzzy logic inference operation on the extracted candidate preference channel or candidate preference program and calculating a second probability value for each candidate preference channel or candidate preference program as a result of the fuzzy logic inference operation; And
Determining a preference channel or a preference program according to the viewing pattern information of the user based on the calculated first probability value and the second probability value;
Method for recommending a preferred channel / program comprising a. 10. The method of claim 9,
The extracting step,
Apply an entropy function to a first probability value calculated from the Bayesian network learner,
If the result of the entropy function is less than or equal to a predetermined threshold as a result of the application, part or all of the channel or program corresponding to the first probability value is extracted as the candidate preferred channel or candidate preferred program. How to recommend the program. 10. The method of claim 9,
Computing the second probability value,
A fuzzy logic operation is performed on each of the candidate preference channels or candidate preference programs based on preferences and preset rules received from a user;
And a second probability value for each of the candidate preferred channels or programs as a result of the performing. 12. The method of claim 11,
Computing the second probability value,
Performing fuzzy to give a degree of belonging to the preference input from the user and outputting a fuzzy value as a result of performing the fuzzy;
Performing fuzzy inference based on the output fuzzy values and predetermined rules; And
Performing a dispersing on the result value obtained as a result of the fuzzy inference and outputting a second probability value which is unpurged as the result;
Method for recommending a preferred channel / program comprising a. 12. The method of claim 11,
The preference is,
Method for recommending a preferred channel / program, characterized in that the information indicating the preferences or preferences for each channel or program received from the user. 12. The method of claim 11,
The preset rules refer to an IF-THEN rule, which is a sentence indicating a relationship between a series of facts. 10. The method of claim 9,
Wherein the determining comprises:
Using the first probability value of the channel or program corresponding to the second probability value among all channels or programs as the second probability value,
A proportional arithmetic operation based on the second probability value is used to update a first probability value of the remaining channel or program, and to determine and recommend a preferred channel or a preferred program based on the updated first probability value. How to recommend. 10. The method of claim 9,
The viewing pattern information,
A method for recommending a preferred channel / program, comprising the amount of TV viewing time per day, the amount of TV watching time per hour, the total viewing time per channel, and the viewing time per program genre.
KR1020120088364A 2011-10-17 2012-08-13 System for recommending favorite channel/program based on tv watching pattern and method thereof KR20130041725A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US13/653,826 US8789109B2 (en) 2011-10-17 2012-10-17 System for recommending favorite channel/program based on TV watching pattern and method thereof

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
KR20110106089 2011-10-17
KR1020110106089 2011-10-17

Publications (1)

Publication Number Publication Date
KR20130041725A true KR20130041725A (en) 2013-04-25

Family

ID=48440808

Family Applications (1)

Application Number Title Priority Date Filing Date
KR1020120088364A KR20130041725A (en) 2011-10-17 2012-08-13 System for recommending favorite channel/program based on tv watching pattern and method thereof

Country Status (1)

Country Link
KR (1) KR20130041725A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10097895B2 (en) 2013-10-11 2018-10-09 Samsung Electronics Co., Ltd Content providing apparatus, system, and method for recommending contents
CN113312511A (en) * 2021-06-11 2021-08-27 北京百度网讯科技有限公司 Method, apparatus, device and computer-readable storage medium for recommending content

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10097895B2 (en) 2013-10-11 2018-10-09 Samsung Electronics Co., Ltd Content providing apparatus, system, and method for recommending contents
CN113312511A (en) * 2021-06-11 2021-08-27 北京百度网讯科技有限公司 Method, apparatus, device and computer-readable storage medium for recommending content
CN113312511B (en) * 2021-06-11 2024-02-27 北京百度网讯科技有限公司 Method, apparatus, device and computer readable storage medium for recommending content

Similar Documents

Publication Publication Date Title
Gupta et al. Multi-armed bandits with correlated arms
Gastner et al. Diffusion-based method for producing density-equalizing maps
US9402101B2 (en) Content presentation method, content presentation device, and program
Buccapatnam et al. Stochastic bandits with side observations on networks
Ferri-García et al. Propensity score adjustment using machine learning classification algorithms to control selection bias in online surveys
Bach et al. Scaling MPE inference for constrained continuous Markov random fields with consensus optimization
Pearl Myth, confusion, and science in causal analysis
Hadi-Vencheh et al. An inverse optimization model for imprecise data envelopment analysis
US8789109B2 (en) System for recommending favorite channel/program based on TV watching pattern and method thereof
Ali et al. New types of dominance based multi-granulation rough sets and their applications in conflict analysis problems
KR20130041725A (en) System for recommending favorite channel/program based on tv watching pattern and method thereof
Rui et al. A regression approach for prediction of Youtube views
CN109348260A (en) A kind of direct broadcasting room recommended method, device, equipment and medium
Xinchang et al. Movie recommendation algorithm using social network analysis to alleviate cold-start problem
Min et al. Granular association rules for multi-valued data
Kamya et al. Fuzzy logic based intelligent question paper generator
Yang et al. Personalized video recommendations for shared accounts
Vannoni A comparative test of the punctuated equilibrium theory: Policy punctuations in tobacco control
Marin et al. Tensor-based collaborative filtering with smooth ratings scale
Utkin et al. Fuzzy decision making using the imprecise Dirichlet model
Torkjazi et al. A fuzzy probabilistic maximum technique to optimize an unconstrained utility based multi objective model
Zolkepli et al. Automatic switching of clustering methods based on fuzzy inference in bibliographic big data retrieval system
Uno et al. A competitive facility location problem on a network with fuzzy random weights
Nguyen et al. An influence analysis of the number of members on the quality of knowledge in a collective
Kang et al. Modeling the user preference of broadcasting content using Bayesian networks

Legal Events

Date Code Title Description
A201 Request for examination
E902 Notification of reason for refusal
E601 Decision to refuse application