KR20240022698A - method of history-based recommendation of a set of Live channels - Google Patents

Abstract

The present invention operates a genetic algorithm by applying the user's history channel information to fitness evaluation and determines the user's preference by applying selection, crossing, and mutation through fitness evaluation of live channels in a set unit rather than preference evaluation in individual channels. It is about a history-based intelligent live channel set recommendation technology that can satisfy the diversity of users while reflecting the According to the present invention, live channels are provided by recommending a small set of live channels (e.g., a set of 8 live channels) suitable for individual subscribers among a large number of live channels (e.g., 200 live channels in total) provided by IPTV operators. It has the advantage of significantly improving user convenience for enjoyment. In particular, while the conventional channel recommendation method performed on an individual channel basis had the problem of recommending a relatively narrow range of channels depending on the viewer's history channel information, according to the present invention, a genetic algorithm is applied to live channels in a set unit. This has the advantage of satisfying both user preferences and diversity.

Description

History-based intelligent live channel set recommendation method {method of history-based recommendation of a set of Live channels}

The present invention generally relates to a technology for recommending live channels using information about live channels previously viewed by a user (history channel information).

In particular, the present invention operates a genetic algorithm by applying the user's history channel information to fitness evaluation, and applies selection, crossbreeding, and mutation through fitness evaluation of live channels on a set basis rather than preference evaluation on an individual channel basis, thereby improving user satisfaction. It is about a history-based intelligent live channel set recommendation technology that can satisfy user diversity while reflecting the preferences of users.

In the case of live channels such as IPTV operators, a large number of channels, approximately 100 to 200 depending on the operator, are provided to subscribers. This is intended to provide rich content, but it also has its drawbacks. The process for users to find the service they want at the current time on a live channel is quite cumbersome. In general, you change channels with a remote control, manually find them by searching the electronic program guide (EPG), or select a list of recently watched channels (history channels).

Recently, in order to reduce this inconvenience, technology that recommends live channels using artificial intelligence (AI) technology or various metadata has been proposed. However, there was a problem that not only did it not properly reflect individual users' channel preferences, but the range of channel recommendations became too narrow.

Republic of Korea Patent No. 10-1642433 (2016.07.19) “Intelligent content recommendation service method considering user preferences” Republic of Korea Patent No. 10-2137887 (2020.07.20) “IPTV movie VOD channel recommendation server and method by combining movie VOD preference information of mobile OTT service of IPTV service” Republic of Korea Patent No. 10-2096474 (2020.03.27) “Content recommendation device and method” Republic of Korea Patent Publication No. 10-2012-0003362 (2012.01.10) “IPTV individual preference program recommendation system based on collaborative filtering algorithm”

The purpose of the present invention is to provide a technology for recommending live channels by utilizing information about live channels that a user has previously viewed (history channel information).

In particular, the purpose of the present invention is to operate a genetic algorithm by applying the user's history channel information to fitness evaluation, and to apply selection, crossing, and mutation through fitness evaluation of live channels on a collective basis rather than preference evaluation on an individual channel basis. By doing so, it provides history-based intelligent live channel set recommendation technology that can reflect user preferences while also satisfying user diversity.

Meanwhile, the problems to be solved by the present invention are not limited to these matters, and other problems to be solved can be understood from the description in this specification.

In order to achieve the above object, the present invention provides a live broadcasting service that provides streaming of a large number of live channels through a network, in which the channel recommendation server 400 provides live broadcasting services including a small number of live channels for each user based on the user's viewing history. We present a method for recommending a set of channels.

The history-based intelligent live channel set recommendation method according to the present invention includes a first step of acquiring history channel information about the live channel viewing history for each user for a certain period of time; A second step of initially generating a channel set population having M preset live channel sets (CS) including n preset live channels; A third step of repeatedly performing generational evolution on the channel set population by performing selection, crossing, and mutation operations by calculating the fitness of the live channel set based on history channel information; A fourth step of selecting a live channel set with good fitness by calculating fitness using history channel information among the M live channel sets included in the channel set population for which generational evolution has been repeatedly performed; A fifth step of setting at least one of the n live channels included in the selected live channel set as a recommendation target.

In the present invention, history channel information may be comprised of a list of live channels watched over a certain period of time for each user, including a channel identifier, channel name, program name, program genre, program time, and ratio of viewing time to running time.

Additionally, in the third step, the fitness F(CS_k) of one live channel set (CS_k) is

It is calculated by summing the weights W(CH_1k), ..., W(CH_nk) of the n live channels (CH_1k ~ CH_nk) that make up the live channel set (CS_k), and at this time, the weight of one live channel (CH_ik) Weight W(CH_ik) is

The channel weight W(t), genre weight W(g), and program weight W(p) for each time slot can be calculated by simple or weighted summation.

Additionally, in the third step, the selection operation may be performed so that a live channel set with a high fitness value has a high probability of being selected based on the fitness F(CS_k) of the M live channel sets included in the channel set population.

Additionally, in the third step, the crossbreeding operation may be performed to create a new live channel set by selecting half of the live channels based on the weight value from each of the two live channel sets selected in the channel set population.

Additionally, in the third step, the mutation operation may be performed by differentially applying the mutation expression rate of the mutation operation so that it is inversely proportional to the weight value of the live channel.

Meanwhile, the computer program according to the present invention is stored in a non-volatile storage medium in order to execute the history-based intelligent live channel set recommendation method on the computer.

According to the present invention, live channels are provided by recommending a small set of live channels (e.g., a set of 8 live channels) suitable for individual subscribers among a large number of live channels (e.g., 200 live channels in total) provided by IPTV operators. It has the advantage of significantly improving user convenience for enjoyment.

In particular, while the conventional channel recommendation method performed on an individual channel basis had the problem of recommending a relatively narrow range of channels depending on the viewer's history channel information, according to the present invention, a genetic algorithm is applied to live channels in a set unit. This has the advantage of satisfying both user preferences and diversity.

[Figure 1] is a diagram showing a live broadcasting service system for the present invention.
[Figure 2] is a flowchart showing a history-based intelligent live channel set recommendation process according to the present invention.
[Figure 3] is an exemplary diagram showing history channel information in the present invention.
[Figure 4] is an example diagram showing a channel set population for the genetic algorithm in the present invention.
[Figure 5] is a conceptual diagram showing the generational evolution of the channel set population by the genetic algorithm in the present invention.
[Figure 6] is an example diagram showing the selection operation of the genetic algorithm according to the fitness ratio in the present invention.
[Figure 7] is a conceptual diagram showing the crossover operation of the genetic algorithm in the present invention.
[Figure 8] is a conceptual diagram showing the mutation operation of the genetic algorithm in the present invention.

Hereinafter, the present invention will be described in detail with reference to the drawings.

[Figure 1] is a diagram showing a live broadcasting service system for the present invention.

The present invention is a technology for recommending live channels to users (subscribers) who receive live broadcasting services based on their viewing history. Referring to [Figure 1], the live broadcasting service system for the present invention is comprised of a plurality of OTT devices 100, a subscriber management server 200, a streaming server 300, and a channel recommendation server 400. [Figure 1] shows a plurality of servers, but they are not limited to being implemented as individual hardware.

First, the OTT device 100 is a device placed in a subscriber space (e.g., living room, store) of a live broadcasting service, and receives a live broadcast stream from the streaming server 300 through a network and plays and outputs it to the subscriber. At this time, the OTT device 100 may be implemented as an OTT terminal (Over-The-Top terminal), a smartphone, an IP set-top box, etc.

The subscriber management server 200 is a device that manages information on live broadcasting service subscribers in the subscriber database 201.

The streaming server 300 is a device that provides live broadcasting services to subscribers. The streaming server 300 manages information about live broadcast channels in the broadcast channel database 301.

The channel recommendation server 400 is a device that selects live channels that can be recommended to subscribers according to the present invention. In other words, it is a device that selects a small number of live channels (e.g. 8) suitable for each subscriber from among the plurality (e.g. 200) live channels provided in the live broadcasting service. As basic information for this, the channel recommendation server 400 manages information about the viewing history of live channels for the past certain period (eg, 2 weeks) for each subscriber in the history database 401.

[Figure 2] is a flowchart showing a history-based intelligent live channel set recommendation process according to the present invention.

The present invention relates to a live broadcasting service system such as [Figure 1] that provides streaming of a plurality of live channels (e.g., 200) through a network, where the channel recommendation server 400 recommends a corresponding user for an individual user (e.g., subscriber 1). For example, it is a technology that recommends a set of live channels including a small number (e.g., 8) live channels based on the viewing history of the past two weeks.

To select live channels, the channel recommendation server 400 uses a genetic algorithm (GA). Genetic algorithm is an optimization technique introduced by John Holland in his 1975 book "Adaptation on Natural and Artificial Systems" and uses the concepts of natural selection and genes, which are the principles of evolution. After randomly generating a solution to a problem, generational evolution consisting of selection, crossover, and mutation is repeatedly performed to find a good solution.

In a genetic algorithm (GA), the solution to a problem is expressed as a fixed-length string, which improves the performance of crossover and mutation operations when generating the next generation. The degree of good or bad for an individual answer is evaluated by an objective function, which is called fitness. The selection operation determines the answer to be used when generating the next generation from the current generation, and the decision is made probabilistically according to the fitness value. The crossover operation combines two answers to create a new answer, simulating the crossover of biological chromosomes. The mutation operation randomly transforms part of the solution, reducing the possibility that the generational evolution process falls into a local optimal solution.

In the present invention, the recommendation range of live channels can be diversified by incorporating a genetic algorithm. The user's preferences are standardized to some extent and are revealed in the viewing history, but at the same time, the preferences may extend to completely strange areas that are not revealed in the viewing history. In the present invention, an approach was adopted to select recommendation targets by inputting information from viewing history into a genetic algorithm. At this time, the viewing history is reflected in the fitness calculation in the genetic algorithm, which affects the selection probability of individual answers during the generational evolution process.

In addition, the present invention adopts a method of applying a genetic algorithm in units of a set of live channels including a plurality of channels (e.g., 8 live channels) rather than a method of obtaining a plurality of preferred channels one by one. . A genetic algorithm is used to select a channel set that is estimated to be the best, and eight live channels included in the selected channel set are recommended to the user.

The history-based intelligent live channel set recommendation process according to the present invention is as follows.

Step (S110): First, history channel information regarding the live channel viewing history for each user for a certain period of time (e.g., the last two weeks) is acquired.

In the present invention, history channel information is viewing information about a list of live channels watched over a certain period of time (e.g., the last two weeks) for each user, such as channel identifier, channel name, program name, program genre, program time, and viewing relative to running time. It is composed including time ratio.

[Figure 3] is an exemplary diagram showing history channel information in the present invention, and history channel information is data reflecting the preferences (tastes) that individual users have for live channels. Preferably, when constructing history channel information, only programs that viewers have watched for more than a certain threshold (e.g., 5 minutes) are reflected.

In the prior art, a list of preferred channels was provided using history channel information, but in this case, there was a problem in that channel recommendations were made only within a limited search space called viewing history.

In the present invention, various operations of a genetic algorithm are performed using history channel information. In other words, the fitness of an individual entity (a set of live channels) is calculated using the history channel information, and the selection operation, crossover operation, and mutation operation of the genetic algorithm are performed based on the calculated fitness. As a result, history channel information indirectly affects generational evolution by genetic algorithm.

Step (S120): Next, a channel set population having M (e.g., 20) live channel sets (CS1 to CS20) including n (e.g., 8) live channels is initially created.

[Figure 4] is an exemplary diagram showing the channel set population for the genetic algorithm in the present invention. In [Figure 4], the channel set population consists of 20 live channel sets (CS1 to CS20), and one live channel set (CS_k) consists of 8 live channels. As generation evolution progresses according to the genetic algorithm, the channel set population also changes, and a generation number can be added to the channel set population to reflect this. In genetic algorithms, initial conditions are generally not important. Accordingly, when initially generating a channel set population, the identifiers (IDs) of n live channels included in each live channel set (CS_k) can be randomly set.

Step (S130): Next, generational evolution is repeatedly performed on the channel set population by performing selection, crossing, and mutation operations by calculating the fitness of the live channel set based on the history channel information.

In step S130, generational evolution is repeatedly performed from the channel set population initially created in step S120 through the selection operation, crossover operation, and mutation operation of the genetic algorithm. [Figure 5] is a conceptual diagram showing the generational evolution of the channel set population by the genetic algorithm in the present invention.

Through this generational evolution, the components (live channels) of the live channel set included in the channel set population gradually change, and the factor that determines the direction of change is fitness. Fitness determines the probability that each set of live channels will survive selection and crossing operations and survive into the next generation. In other words, a solution with high fitness can become the dominant species in the process of generational evolution. In the present invention, suitability is calculated based on the user's history channel information.

In the present invention, the process of calculating suitability for a set of live channels is described.

In the present invention, suitability is calculated in units of live channel sets (CS). One live channel set (CS_k) includes a plurality of live channels, that is, n (e.g. 8) live channels (CH_1k to CH_nk), and the fitness F (CS_k) of the live channel set (CS_k) is It is calculated by adding up the weights W(CH_1k), ..., W(CH_nk) of the channels (CH_1k to CH_nk).

At this time, the weight for the live channel (CH_ik) is calculated from the history channel information of the user. In a preferred embodiment, the weight W(CH_ik) of one live channel (CH_ik) is calculated by simple or weighted summation of the channel weight W(t), genre weight W(g), and program weight W(p) for each time slot. do.

At this time, the channel weight W(t) for each time zone represents the user's preference (e.g., number of views, share of viewing time) for a specific channel shown at a specific time in the history channel information. The genre weight W(g) represents the user's preference for a specific genre (e.g., distribution of viewing counts) in the history channel information. The program weight W(p) represents the preference for a specific program indicated by the user in the history channel information. If the program name that the viewer previously watched is the same or similar, the weight is increased.

From this definition, the weight for a specific live channel (CH_i) is a value that changes over time. The channel weight W(t) for each time slot changes depending on the user's viewing pattern over time as shown in the history channel information. The genre weight W(g) and program weight W(p) change in response to the genre and name of the program being transmitted or scheduled to be transmitted according to the program schedule of each live channel (CH_i).

Next, the selection operation of the genetic algorithm is described.

The selection operation is an operation that determines the solution (live channel set) that will survive when generating the channel set population of the next generation (gn = p+1) from the channel set population of the current generation (gn = p). In general, a selection operation selects fewer answers than the entire population. For example, the channel set population [1] consists of 20 live channel sets, and through a selection operation, for example, 10 live channel sets are selected and survive in the channel set population [2].

For the 20 live channel sets (CS1 to CS20) included in the channel set population [1], the fitness F(CS_1) to F(CS_20) is calculated according to [Equation 1] and [Equation 2], and the selection operation is is made based on the fitness F(CS_k) of each live channel set.

As an example, the 10 live channel sets with the highest fitness values may be selected among the 20 live channel sets (CS1 to CS20) included in the channel set population [1].

As another example, the selection of 20 live channel sets (CS1 to CS20) included in the channel set population [1] may be determined probabilistically according to their respective suitability. For example, as shown in [FIG. 6], a roulette selection range may be set according to each suitability ratio, a random number may be generated, and a set of 10 live channels may be selected based on the results.

Next, the crossover operation of the genetic algorithm is described.

The crossover operation is a process of combining two answers (live channel sets) to create a new answer (live channel set), and conceptually, it simulates the chromosome crossover of a living organism. Preferably, two live channel sets can be selected from the ten surviving live channel sets through a selection operation and combined to create a new live channel set. At this time, the process of selecting two from a set of 10 live channels can also be configured to correspond to each suitability. Depending on the implementation, since the fitness has already been reflected in the selection operation, the mating operation may be configured to select randomly without reflecting the fitness again.

[Figure 7] is a conceptual diagram showing the crossover operation of the genetic algorithm in the present invention.

Among the 10 live channel sets constituting the channel set population, two live channel sets CS1 and CS7 were selected for crossbreeding. Select half (e.g., four) live channels from each of CS1 and CS7 to create a new set of live channels. In [Figure 7], {CH10, CH81, CH214, CH7} was selected from CS1, and {CH155, CH131, CH9, CH22} was selected from CS7. In this way, in the process of selecting four channels from the live channel set, it is desirable to base these channels on the weights calculated by [Equation 2]. That is, an embodiment in which the top four channels are selected according to the order of weights is possible, and an embodiment in which selection is probabilistically determined corresponding to each weight is also possible.

When generating the channel set population of the next generation (gn = p+1) from the channel set population of the current generation (gn = p), some (e.g., 10) live channel sets were determined through a selection operation. The remaining solutions to form the channel set population (e.g., a set of 10 live channels) are obtained through a crossover operation, which maintains the size of the channel set population.

Next, the mutation operation of the genetic algorithm is described.

The mutation operation randomly transforms a part of the solution (live channel set) and reduces the possibility that the channel set population falls into a local optimal solution through a genetic algorithm. In the mutation operation, a small number of live channels are randomly selected from the channel set population and changed to completely different channel values.

[Figure 8] is a conceptual diagram showing the mutation operation of the genetic algorithm in the present invention.

Since the channel set population consists of 20 live channel sets, and one live channel set consists of 8 live channels, the channel set population contains a total of 160 live channels. If the mutation expression rate is applied at 2.5%, a total of 4 live channels are randomly selected for the channel set population and changed to different channel values.

Depending on the implementation example, the mutation expression rate of the mutation operation may be differentially applied in response to the weight value of the live channel (so that it is inversely proportional to the weight value). In other words, a relatively low mutation expression rate is applied to live channels with a high weight, and a relatively high mutation expression rate is applied to live channels with a low weight. As an example, all live channels (eg, 160) included in the channel set population are classified into a first group with a low weight and a second group with a high weight. Classification can be based on the results of comparison with a preset threshold or by weight ranking. A mutation operation may be performed by applying a high mutation expression rate to the first group of live channels and a low mutation expression rate to the second group of live channels. For example, a 10-fold mutation expression rate is applied to the first group compared to the second group. This works to strengthen the user's preference while applying mutation operation for diversity of channel recommendations.

Steps (S140, S150): Next, the history channel among M (e.g. 20) live channel sets (CS1 [80] to CS20 [80]) included in the channel set population after repeated generation evolution has been performed. By calculating the suitability using information, a set of live channels with good (highest) suitability (e.g. CS1 [80]) is selected. Then, n (e.g., 8) live channels included in the selected live channel set (e.g., CS1 [80]) are set as recommendation targets for the user.

In general, generational evolution of genetic algorithms is known to be effective when approximately 80 to 90 generations are performed. As generational evolution progresses, the number of live channel sets with high suitability as a solution increases within the channel set population. When the process progresses beyond a certain point, the degree of suitability converges and there is no need to proceed with generational evolution any longer. Accordingly, generational evolution is usually performed only up to a preset number of times, and [Figure 5] shows generational evolution performed 79 times.

In [Figure 5], the result of performing generational evolution according to the genetic algorithm 79 times is the channel set population [gn = 80]. This channel set population [gn = 80] includes 20 live channel sets (CS1 [80] to CS20 [80]), of which the fitness values calculated according to [Equation 1] and [Equation 2] Select this set of good live channels. A set of live channels with a suitability greater than a preset threshold can be selected, or a set of live channels with the highest suitability value can be selected. Assume that the first live channel set CS1 [80] is selected. This CS1 [80] includes 8 live channels, and all or some of them (selecting those with a weight greater than a threshold) are set as recommendations to the user.

Meanwhile, the present invention can be implemented in the form of computer-readable code on a computer-readable non-volatile recording medium. These non-volatile recording media include various types of storage devices, such as hard disks, SSDs, CD-ROMs, NAS, magnetic tapes, web disks, and cloud disks. Additionally, the present invention can be implemented in a form in which code is distributed and stored and executed in a plurality of storage devices connected through a network. Additionally, the present invention may be implemented in the form of a computer program stored on a medium in order to execute a specific procedure in combination with hardware.

100: OTT device
200: Subscriber management server
201: Subscriber database
300: Streaming server
301: Broadcast channel database
400: Channel recommendation server
401: History database

Claims (7)

In a live broadcasting service that provides streaming of multiple live channels through a network, the channel recommendation server 400 recommends a set of live channels including a small number of live channels to an individual user based on the user's viewing history, comprising:
A first step of acquiring history channel information about the live channel viewing history for each user over a certain period of time;
A second step of initially generating a channel set population having M preset live channel sets (CS) including n preset live channels;
A third step of repeatedly performing generational evolution on the channel set population by performing a selection operation, a crossing operation, and a mutation operation by calculating the fitness of the live channel set based on the history channel information;
A fourth step of selecting a live channel set with good suitability by calculating suitability using the history channel information from among the M live channel sets included in the channel set population for which the generational evolution has been repeatedly performed;
A fifth step of setting at least one of the n live channels included in the selected live channel set as a recommendation target;
A history-based intelligent live channel set recommendation method comprising:
In claim 1,
The history channel information is history-based, which consists of a list of live channels watched over a certain period of time for each user, including channel identifier, channel name, program name, program genre, program time, and ratio of viewing time to running time. Intelligent live channel set recommendation method.
In claim 1,
In the third step, the fitness F(CS_k) of one live channel set (CS_k) is

Calculated by adding up the weights W(CH_1k), ..., W(CH_nk) of the n live channels (CH_1k to CH_nk) constituting the live channel set (CS_k),
At this time, the weight W(CH_ik) of one live channel (CH_ik) is

A history-based intelligent live channel set recommendation method characterized by calculating the channel weight W(t), genre weight W(g), and program weight W(p) for each time slot by simple or weighted summation.
In claim 3,
In the third step, the selection operation is performed so that a live channel set with a high fitness value has a high probability of being selected based on the fitness F(CS_k) of the M live channel sets included in the channel set population. A history-based intelligent live channel set recommendation method.
In claim 4,
In the third step, the crossbreeding operation is performed to generate a new live channel set by selecting half of the live channels based on the weight value from each of the two live channel sets selected in the channel set population. Intelligent live channel set recommendation method.
In claim 5,
In the third step, the mutation operation is performed by differentially applying the mutation expression rate of the mutation operation so as to be inversely proportional to the weight value of the live channel.
A computer program stored in a storage medium to execute the history-based intelligent live channel set recommendation method according to any one of claims 1 to 6 on a computer.

Images