KR20210097432A - Apparatus and method for recommending contents - Google Patents

Abstract

Disclosed are a device and a method for recommending a content, which uniquely recommends the content for each user by reflecting the characteristics of each user at a low cost and fast time. According to an embodiment of the present invention, a user terminal for recommending the content to the user includes: a communication unit communicating with a content service server and receiving a user's offset recommendation model derived from an integrated recommendation model learned from a content service server; an input unit for receiving an input for using content from the user; a service unit for requesting, receiving, and outputting the content to the content service server through the communication unit based on the input; and a recommendation unit which updates the offset recommendation model based on a user's content use history and recommends the content to the user through the updated offset recommendation model.

Description

Apparatus and method for recommending contents

The present invention relates to an apparatus and method for recommending content, and more particularly, to an apparatus and method for quickly recommending content in various user terminal environments.

In general, users who consume content find and use the content they want through a search. This method has a limitation in efficiently and accurately searching for a content desired by a user in an environment in which the number of content increases exponentially. As an alternative to this, a technology for automatically recommending content that may be of interest to a user has been proposed in consideration of the user's personal information or content use history. It is a kind of target marketing strategy.

However, the conventional content recommendation technology requires a lot of time and computing resources to analyze the user's content use history when the number of users increases. It is convenient from the user's point of view, but there is a problem in that it takes a lot of time and money from the point of view of a content service provider. In addition, since the conventional content recommendation technology recommends content based on the content usage history of all users, it is often not possible to reflect the characteristics of each user.

Some recommend content using a neural network-based content recommendation model such as deep learning. However, it takes a lot of time to learn and model the content recommendation model for all users, so only limited learning is performed. Although some users are grouped and recommendations are provided for each user group, this may not reflect the characteristics of each user. Therefore, there is a need for a method that can reduce time and cost in content creation by reflecting the characteristics of each user. In particular, with the development of broadband Internet and mobile communication technology, especially with the introduction of 5G services, OTT (Over The Top) services that allow users to use media contents such as broadcast programs and movies through various user terminals such as PCs and smartphones are expanding. The need for it is growing even more.

The present invention has been proposed to solve the above problems, and it is an object of the present invention to provide an apparatus and method capable of uniquely recommending content for each user by reflecting the characteristics of each user at a low cost and fast time.

A user terminal for recommending content to a user according to an embodiment includes: a communication unit communicating with a content service server and receiving an offset recommendation model of the user derived from an integrated recommendation model learned from the content service server; an input unit for receiving an input for using content from the user; a service unit for requesting, receiving, and outputting contents to the contents service server through the communication unit based on the input; and a recommendation unit that updates the offset recommendation model based on the user's content use history and recommends content to the user through the updated offset recommendation model.

The communication unit receives a content recommendation list recommended by the integrated recommendation model from the content service server, and the recommendation unit generates a content recommendation list unique to the user based on the content recommendation list using the offset recommendation model. can create

The recommendation unit may generate a content recommendation list unique to the user by rearranging the received content recommendation list using the offset recommendation model.

The offset recommendation model received by the communication unit may be generated by the content service server derived from the integrated recommendation model based on a difference between a content recommendation result of the integrated recommendation model and a content use history of the user.

The content use history may include at least one of a user's speech, viewing, purchase, or star rating history.

As the user's viewing history, a vector value for each content may be calculated by Equation 1 or Equation 2 below and used for learning.

(Equation 1)

Viewing history vector value for each content = 1+log (total watch time/running time of that content)

(Equation 2)

Viewing history vector value for each content = 1+log(total watch time/running time of that content)+log(α/number of views). where α=1 or determined proportional to the running time of the content.

As the user's purchase history, a standardized vector value may be calculated and used for learning by using the user's purchase amount for each content and the average amount and variance of all users who purchased each content.

As the user's star point history, a standardized vector value may be calculated and used for learning by using the star point given by the user for each content and the average and variance of the star point given by all users for each content.

The user terminal may further include a sharing unit that transmits and shares the updated offset recommendation model to another user terminal.

The sharing unit may transmit and share the content usage history of the user to the other user terminal based on the computing environment of the user terminal.

The sharing unit may stop updating the offset recommendation model of the other user terminal by notifying the other user terminal that the update is scheduled before updating the offset recommendation model.

According to an embodiment, a method for recommending content to a user in a user terminal includes: receiving an offset recommendation model of the user derived from an integrated recommendation model learned from a content service server from the content service server; receiving an input for using content from the user; requesting, receiving, and outputting content from the content service server based on the input; and updating the offset recommendation model based on the content usage history of the user and recommending content to the user through the updated offset recommendation model.

The method further includes receiving, from the content service server, a content recommendation list recommended by the integrated recommendation model, wherein the recommending includes: the user based on the content recommendation list using the offset recommendation model You can create your own content recommendation list.

The recommending may include rearranging the received content recommendation list using the offset recommendation model to generate a content recommendation list unique to the user.

The offset recommendation model received in the receiving step may be generated by the content service server derived from the integrated recommendation model based on a difference between the content recommendation result of the integrated recommendation model and the user's content use history.

The content use history may include at least one of a user's speech, viewing, purchase, or star rating history.

As a user's viewing history, a vector value for each content may be calculated by Equation 3 or Equation 4 below and used for learning.

(Equation 3)

Viewing history vector value for each content = 1+log (total watch time/running time of that content)

(Equation 4)

Viewing history vector value for each content = 1+log(total watch time/running time of that content)+log(α/number of views). where α=1 or determined proportional to the running time of the content.

As the user's purchase history, a standardized vector value may be calculated and used for learning by using the user's purchase amount for each content and the average amount and variance of all users who purchased each content.

As the user's star point history, a standardized vector value may be calculated and used for learning by using the star point given by the user for each content and the average and variance of the star point given by all users for each content.

The method may further include transmitting and sharing the updated offset recommendation model to another user terminal.

The method may further include, before updating the offset recommendation model, stopping the update of the offset recommendation model of the other user terminal by notifying the other user terminal that the update is scheduled.

In the present invention, after modeling an integrated recommendation model for all users in a content service server, an offset recommendation model is generated for each user based on each user's individual content use history and transmitted to each user terminal. By recommending content through the offset recommendation model, the terminal distributes computing resources required to recommend content, and can recommend content suitable for individual characteristics in a short time and at low cost.

According to the present invention, since the offset recommendation model recommends content suitable for the user's characteristics within the range of the content recommendation list recommended by the integrated recommendation model in each user terminal, the offset recommendation model can be produced with a lighter weight than the integrated recommendation model. Since a lot of calculations are not performed in a user terminal in a bad environment, it is efficient to use resources and time required for content recommendation can be saved.

The present invention allows a user to share an offset recommendation model updated in a specific user terminal with other user terminals when a user uses a plurality of user terminals, so that the same content service can be used consistently in a plurality of user terminals according to individual characteristics. let it be

1 is a diagram illustrating a content service system according to an embodiment of the present invention.
2 is a diagram showing the configuration of a content service server according to an embodiment of the present invention.
3 is a diagram showing the configuration of a user terminal according to an embodiment of the present invention.
4 is a flowchart illustrating a method of recommending content in a content service system according to an embodiment of the present invention.

The above-described objects, features, and advantages will become more apparent through the following detailed description in relation to the accompanying drawings, whereby those of ordinary skill in the art to which the present invention pertains can easily implement the technical idea of the present invention. There will be. In addition, in the description of the present invention, if it is determined that a detailed description of a known technology related to the present invention may unnecessarily obscure the gist of the present invention, the detailed description thereof will be omitted. Hereinafter, a preferred embodiment according to the present invention will be described in detail with reference to the accompanying drawings.

1 is a diagram illustrating a content service system according to an embodiment of the present invention. Referring to FIG. 1 , the content service system according to the present embodiment includes a content service server 110 , a user terminal 120 , and a communication network 130 connecting them. Here, the communication network 130 includes, but is not limited to, a known mobile communication network or a next-generation mobile communication network, a wired Internet, or short-range wireless communication, and is not particularly limited as long as it is a network for wired and wireless communication.

The content service server 110 is a service equipment that transmits content to the user terminal 120 according to the request of the user terminal 120 . Here, the content may include, but is not limited to, video content such as a real-time broadcast program or VOD broadcast program or movie, sound source content, and the like. In the case of a real-time broadcasting program, the content service server 110 may receive image content from a broadcasting system of a broadcasting station and stream and transmit it to the user terminal 120 in real time. Upon request of 120 , streaming may be transmitted to the user terminal 120 .

The content service server 110 generates a neural network-based integrated recommendation model for all users, and generates an offset recommendation model for each user based on this. The content service server 110 transmits the content recommendation list generated based on the integrated recommendation model to the user terminal 120 and transmits an offset recommendation model for the user. Preferably, the integrated recommendation model may be updated every first cycle in the content service server 110 . In addition, the offset recommendation model for each user may be generated for each user at the second cycle in the content service server 110 and transmitted to each user terminal 120 . In this case, the second period may be equal to or longer than the first period. For example, the first cycle may be 1 day, and the second cycle may be a longer period than the first cycle, such as 1 day or 1 week. Meanwhile, after the offset recommendation model for each user is initially generated by the content service server 110 , it may be updated in the user terminal 120 until a new offset recommendation model is received from the content service server 110 .

The user terminal 120 may access the content service server 110 through the communication network 130 to search for content and select content, or select content from a content recommendation list received from the content service server 110 . can The user terminal 120 receives the content selected by the user from the content service server 110 and reproduces and displays the content on the display device. Preferably, the user terminal 120 stores the offset recommendation model received from the content service server 110 and displays the changed content recommendation order of the content recommendation list received through the offset recommendation model. The user terminal 120 periodically updates the offset recommendation model based on the user's content use history. The user terminal 120 includes a mobile terminal such as a smart phone or a tablet PC, a fixed personal computer (PC), an IPTV set-top box, and an artificial intelligence device such as a GiGA Genie.

2 is a diagram showing the configuration of a content service server according to an embodiment of the present invention. The content service server 110 may include a memory, a memory controller, one or more processors (CPU), a peripheral interface, an input/output (I/O) subsystem, a display device, an input device, and a communication circuit. The memory may include high-speed random access memory, and may also include one or more magnetic disk storage devices, non-volatile memories such as flash memory devices, or other non-volatile semiconductor memory devices. Access to the memory by other components, such as the processor and peripheral interfaces, may be controlled by the memory controller. The memory may store various kinds of information and program instructions, and the program is executed by the processor. Peripheral interfaces connect the I/O peripherals to the processor and memory. One or more processors execute various software programs and/or sets of instructions stored in memory to perform various functions for the system and process data. The I/O subsystem provides an interface between input/output peripherals such as display devices and input devices and the peripheral interface. The communication circuit performs communication through an external port or communication by an RF signal. The communication circuitry converts electrical signals into RF signals and vice versa through which the RF signals can communicate with communication networks, other mobile gateway devices, and communication devices. Referring to FIG. 2 , the content service server 110 includes a membership subscription processing unit 210 , a service providing unit 220 , an integrated recommendation model generation unit 230 , and an offset recommendation model generation unit 240 . These components may be implemented as a program, stored in a memory, and executed by a processor, or may be implemented and operated as a combination of hardware and software.

The membership registration processing unit 210 receives personal information and identification information of the user terminal 120 from a user who intends to use the content service, stores the received information in the storage unit, and processes the membership registration. The membership registration processing unit 210 may receive personal information and identification information of the user terminal 120 from the user terminal 120 to perform membership registration processing. When the user terminal 120 is accessed, the member registration processing unit 210 may perform login authentication using identification information or personal information of the user terminal 120 received from the user terminal 120 . Here, the identification information of the user terminal 120 includes unique identification information such as a phone number or MAC address, and the personal information may include gender, age, name, address, and the like.

The service provider 220 searches for content in the storage unit or transmits content received from an external system to the user terminal 120 according to a request from the user terminal 120 . The service providing unit 220 may provide a search interface for searching for content to the user terminal 120 to receive content selection, or may provide a list of recommended content to the user terminal 120 . Preferably, the service providing unit 220 generates a list of recommended contents by inputting personal information of each user into the integrated recommendation model generated by the integrated recommendation model generation unit 230 to be described below to generate a recommended content list for each user terminal 120 . ), along with each user's offset recommendation model, which will be described below.

The integrated recommendation model generator 230 generates an integrated recommendation model applied to all users based on a neural network such as deep learning. The integrated recommendation model generator 230 generates an integrated recommendation model by learning the neural network using personal information of all users and content use histories of all users as learning data. An example of the neural network may include, but is not limited to, a convolutional neural network (CNN). Here, the content use history may include at least one of user's speech, viewing, purchase, and star rating information. Here, the personal information may include personal preference information extracted through the content use history. For example, the personal preference information may be a preferred genre, preferred person (director, actor, etc.), preferred topic, preferred temporal/spatial background, etc. extracted through the user's content use history. Accordingly, the integrated recommendation model generator 230 may extract personal preference information through the content use history.

Specifically, the integrated recommendation model generation unit 230 collects the content use history of all users for N days (N is a natural number), and uses the collected content use history and personal information of all users as learning data. It is fed into the neural network to generate a unified recommendation model. For example, by collecting the content usage history used by User 1 for N days, and inputting the personal preference information of User 1 and the content usage history of the corresponding User 1 to the neural network, personal preference information and contents of all users The neural network is learned by inputting the usage history into the neural network. Therefore, a neural network-based integrated recommendation model that can be commonly applied to all users is created. In general, a neural network is composed of three layers: an input layer, a hidden layer, and an output layer, and each layer is composed of nodes. The unified recommendation model can be understood as a weight between each node of such a neural network.

When a specific user's personal information, preferably personal preference information, is input into the learned integrated recommendation model, the integrated recommendation model generates a recommended content list using the specific user's personal information. Since the integrated recommendation model is learned using the personal information and content use histories of all users, it is possible to find contents that the specific user may like based on the contents usage histories used by other users who have similar personal information to the specific user's personal information. A list is recommended. Since the integrated recommendation model is learned using the data of all users, it does not fully reflect the characteristics of each user, and it takes a lot of time to learn. Therefore, it is preferable to learn using data of a certain number of users in learning the integrated recommendation model.

The integrated recommendation model generator 230 vectorizes the content usage history input as learning data in generating the integrated recommendation model and inputs it to the neural network. In this case, the chronological order of the content use history is ignored. The reason that the time order is not considered is because the user's content preference information is considered as the most important factor rather than the order. Hereinafter, a process of vectorizing each of the viewing history, purchase history, utterance history, and star rating history will be described as a specific example.

First, the content viewing history is calculated as a vector value using the running time of the content viewed by the user, the number of times the content is viewed, and the viewing time. For example, it is assumed that the running times of Content 1, Content 2, and Content 3 are as follows [Table 1].

division Content 1 content 2 content 2 running time 60 20 40

In an embodiment, the viewing history vector value for each content may be calculated as (1+log(total viewing time/running time of the corresponding content)). For example, if user A watches content 1 for 30 minutes, watches content 2 for 20 minutes, and does not watch content 3, user A's vector values for content 1, content 2, and content 3 are (0.699, 1, 0). When user B watches content 1 for 10 minutes, content 2 for 10 minutes, and content 3 for 10 minutes, user B's vector values for content 1, content 2, and content 3 are (0.222, 0.699, 0.397) ) becomes

In another embodiment, the viewing history vector value for each content may be calculated as (1+log(total viewing time/running time of the corresponding content)+log(1/number of views)). At this time, if it becomes a negative number, it is set to 0. For example, if user A watches content 1 once for a total of 30 minutes, content 2 twice for a total of 20 minutes, and does not watch content 3, user A's The vector value becomes (0.699, 0.699, 0). If user B views Content 1 once for a total of 10 minutes, Content 2 twice for a total of 10 minutes, and Content 3 twice for a total of 10 minutes, User B's The vector value becomes (0.222, 0.397, 0).

In another embodiment, the viewing history vector value for each content may be calculated as (1+log(total viewing time/running time of the corresponding content)+log(α/number of views)). At this time, if it becomes a negative number, it is set to 0, and α is determined in proportion to the running time of the content. For example, when the length of the content is less than 10 minutes, α=1, when the length of the content is 10 to 30 minutes, α=2, and when it exceeds 30 minutes, α=3.

Next, the content purchase history may calculate a vector value based on whether the user purchases each content and the purchase amount. In an embodiment, a vector value for the specific content is 1 when the specific content is purchased, and 0 when the specific content is not purchased. In another embodiment, a value obtained by normalizing the Z-Score through the amount of content purchased by the user and the total average amount and variance of all users who purchased the content may be a vector value of each content.

Next, for the content star rating history, a vector value may be calculated based on the user's star rating for each content (ie, a score given by a method such as 1 to 4 stars as a score). In an embodiment, the star score given by the user for each content may be used as a vector value as it is. In another embodiment, a Z-Score normalization value obtained by normalizing the star points given by the user to each content through the overall average and variance of the star points given by all users to each content may be a vector value of each content.

Finally, the content utterance history may be a content search/recommendation history among conversation histories uttered by a user in an artificial intelligence device such as GiGA Genie or a mobile terminal that provides a voice recognition service such as a smart phone. Here, the conversation history includes not only content search in the content service server 110, but also content search/recommendation in other content servers, for example, YouTube, content search/recommendation for using a sound source server such as Genie Music, etc. do. The number of utterances of the user for each content may be calculated as a vector value as it is.

The integrated recommendation model generation unit 230 generates an integrated recommendation model applied to all users based on a neural network such as deep learning, in addition to the digitized vector values of viewing, purchase, star rating, and speech history, the content used by the user You can also use more metadata. In addition, the integrated recommendation model generator 230 may generate a recommendation model for each type of user terminal 120 of all users (eg, set-top box, mobile terminal, etc.) and merge them to create an integrated recommendation model, , or by each type of content use history, that is, a viewing history, a purchase history, a star rating history, and a speech history, respectively, a recommendation model may be generated and merged to create an integrated recommendation model.

The offset recommendation model generator 240 generates an offset recommendation model for each user derived from the integrated recommendation model. Specifically, the offset recommendation model generation unit 240 calculates a difference between a content recommendation result for each user through the integrated recommendation model and a content use history for each user, and reflects the difference to each user from the integrated recommendation model. Create a star offset recommendation model. That is, the offset recommendation model of each user is generated from the integrated recommendation model based on how different the viewing, purchasing, star rating, and speech history for each user is from the average user. In this case, the difference means a distance difference or similarity between vectors of content use histories. The offset recommendation model generator 240 collects a history for M (M is a natural number) days in collecting each user's content use history. The offset recommendation model generator 240 generates an offset recommendation model only once from the integrated recommendation model for each user.

When the service provider 220 generates a content recommendation list for each user using an offset recommendation model for each user whenever users use a service, a lot of computing resources are required, and thus a lot of time is required to recommend content. This takes However, as in the present embodiment, the content service server 110 generates only an average content recommendation list through an integrated recommendation model commonly used for all users, and the content recommendation calculation suitable for individual characteristics of each user is performed for each user. It is made in the terminal 120 and has the same effect as distributing computing resources for content recommendation to each user, so that content suitable for individual characteristics can be quickly recommended.

3 is a diagram showing the configuration of a user terminal according to an embodiment of the present invention. The user terminal 120 may include a memory, a memory controller, one or more processors (CPU), a peripheral interface, an input/output (I/O) subsystem, a display device, an input device, and a communication circuit. The memory may include high-speed random access memory, and may also include one or more magnetic disk storage devices, non-volatile memories such as flash memory devices, or other non-volatile semiconductor memory devices. Access to the memory by other components, such as the processor and peripheral interfaces, may be controlled by the memory controller. The memory may store various kinds of information and program instructions, and the program is executed by the processor. Peripheral interfaces connect the I/O peripherals to the processor and memory. One or more processors execute various software programs and/or sets of instructions stored in memory to perform various functions for the system and process data. The I/O subsystem provides an interface between input/output peripherals such as display devices and input devices and the peripheral interface. The communication circuit performs communication through an external port or communication by an RF signal. The communication circuitry converts electrical signals into RF signals and vice versa through which the RF signals can communicate with communication networks, other mobile gateway devices, and communication devices. Referring to FIG. 3 , the user terminal 120 includes a communication unit 310 , an input unit 320 , a service unit 330 , a recommendation unit 340 , and a sharing unit 350 . These components may be implemented as a program, stored in a memory, and executed by a processor, or may be implemented and operated as a combination of hardware and software.

The communication unit 310 communicates with the content service server 110 to transmit and receive various data. The communication unit 310 receives the user's offset recommendation model derived from the integrated recommendation model learned from the content service server 110 . In addition, the communication unit 310 receives the content recommendation list generated from the integrated recommendation model from the content service server 110 together when receiving the offset recommendation model, and also periodically receives the integrated recommendation model from the content service server 110 . Receive a content recommendation list created in .

The input unit 320 receives an input for using content from an input device, for example, a mouse, a touch screen, a remote control, or a microphone. For example, a selection input of a specific content from the content recommendation list may be received by voice or through a remote controller. Alternatively, the input unit 320 may receive a keyword for content search from the user through the search input window received from the content service server 110 , or may receive an input for menu search.

The service unit 330 receives data from the content service server 110 through the communication unit 310 and displays it on the display device. Various user interfaces (UIs) can be displayed, and content streamed from the content service server 110 is reproduced and displayed on the display device. The service unit 330 requests content from the content service server 110 through the communication unit 310 based on the user input received through the input unit 320 and receives the corresponding content.

The recommendation unit 340 stores the user's offset recommendation model received from the content service server 110 through the communication unit 310 in the storage unit, generates a content recommendation list based on the offset recommendation model, and the service unit It is transmitted to 330 to be displayed on the display device. Preferably, the recommendation unit 340 filters and rearranges the content recommendation list generated by the integrated recommendation model received from the content service server 110 based on the offset recommendation model. The recommendation unit 340 inputs the content recommendation list to an offset recommendation model, and the offset recommendation model re-calculates the ranking of each content in the content recommendation list to rearrange the contents. In this case, the recommendation unit 340 may exclude content having a rank lower than the threshold among the reordered content ranks from the recommendation target.

The recommendation unit 340 stores the user's content use history input through the input unit 320 in the storage unit, and updates the offset recommendation model based on the content use history. The offset recommendation model is already learned by using the user's content use history once in the content service server 110 , and the recommender 340 uses the user's new content use history in the user terminal 120 additionally. Update the offset recommendation model. The recommendation unit 340 may update the offset recommendation model at regular intervals, or may update the offset recommendation model according to a user's input.

The sharing unit 350 may store information of the user terminals 120 owned by the user in the storage unit, and transmit and share the offset recommendation model to other user terminals 120 possessed by the user. The sharing unit 350 may periodically share status information with other user terminals 120 . Here, the state information may include an update time of the offset recommendation model, a computing environment (eg, CPU usage rate, memory usage rate, etc.). When the offset recommendation model is updated, the sharing unit 350 transmits and shares it to another user terminal 120 .

Sharing unit 350, when the computing environment of the user terminal 120 is not good, for example, when the load of the CPU is above a threshold, the content usage history is transmitted to the other user terminal 120 having a good computing environment and the corresponding It can be used to update the offset recommendation model in other user terminals 120 . Also, before the recommendation unit 340 starts updating the offset recommendation model, the sharing unit 350 notifies the other user terminal 120 that the update is scheduled, and recommends the updated offset to the other user terminal 120 . The update of the offset recommendation model may be stopped until the model is shared.

4 is a flowchart illustrating a method of recommending content in a content service system according to an embodiment of the present invention.

Referring to FIG. 4 , in step S401 , the content service server 110 generates an integrated recommendation model applied to all users based on a neural network such as deep learning. The content service server 110 generates an integrated recommendation model by learning the neural network using personal information of all users and content use histories of all users as learning data. An example of the neural network may include, but is not limited to, a convolutional neural network (CNN). Here, the content use history is at least one of the user's speech, viewing, purchase, and star rating information. Here, the personal information may include personal preference information extracted through the content use history. For example, the personal preference information may be a preferred genre, preferred person (director, actor, etc.), preferred topic, preferred temporal/spatial background, etc. extracted through the user's content use history.

Specifically, the content service server 110 collects the content use history of all users for N days (N is a natural number), and stores the collected content use history and personal information of all users as learning data in the neural network. Input to create an integrated recommendation model. For example, by collecting the content usage history that User 1 used for N days, and inputting the personal preference information, which is the personal information of User 1, and the content use history of the corresponding User 1 to the neural network, the personal information of all users and content usage history are input into the neural network to learn the neural network. Therefore, a neural network-based integrated recommendation model that can be commonly applied to all users is created. In general, a neural network is composed of three layers: an input layer, a hidden layer, and an output layer, and each layer is composed of nodes. The unified recommendation model can be understood as a weight between each node of such a neural network.

In step S402, the content service server 110 inputs each user's personal information into the learned integrated recommendation model to generate a recommended content list for each user. Since the integrated recommendation model is trained using all users' personal information and content use histories, a list of content that each user may like is generated based on the content use histories used by other users who have personal information similar to each user's personal information. it will be recommended

In step S403, the content service server 110 generates an offset recommendation model for each user derived from the integrated recommendation model. Specifically, the content service server 110 calculates the difference between the content recommendation result for each user and the content use history for each user through the integrated recommendation model, and reflects the difference to the offset for each user from the integrated recommendation model. Create a recommendation model. That is, the offset recommendation model of each user is generated from the integrated recommendation model based on how different the viewing, purchasing, star rating, and speech histories for each user are from the average user. In this case, the difference means a distance difference or similarity between vectors of content use histories. The content service server 110 collects the history for M (M is a natural number) days in collecting the content use history of each user. The offset recommendation model generator 240 generates an offset recommendation model only once from the integrated recommendation model for each user.

In steps S404 and S405 , the content service server 110 transmits the content recommendation list for each user generated in step S402 and the offset recommendation model for each user generated in step S403 to each user terminal 120 .

In step S406, the user terminal 1 120 stores the received offset recommendation model in the storage unit, and recommends content to the user by filtering and rearranging the received content recommendation list based on the offset recommendation model. The user terminal 1 120 inputs the content recommendation list into an offset recommendation model, and the offset recommendation model re-calculates the rank of each content in the content recommendation list to rearrange the contents. In this case, the user terminal 1 120 may exclude content having a rank lower than the threshold among the reordered contents from the recommendation target.

In step S407, the user terminal 1 120 may request the content from the content service server 110 according to the user's input, and receive the received content accordingly. The user terminal 1 120 stores the user's content use history in the storage unit.

In step S408, when a predetermined time elapses, the user terminal 1 120 updates the offset recommendation model based on the user's content use history stored in the storage unit. The offset recommendation model has already been learned by the content service server 110 once using the user's content use history, and the user terminal 1 120 additionally updates the offset recommendation model using the user's new content use history. will do The user terminal 1 120 may update the offset recommendation model at regular intervals, or may update the offset recommendation model according to a user's input.

In step S409, user terminal 1 120 may transmit and share the updated offset recommendation model to another user terminal n 120 owned by the user. In another embodiment, the user terminals 120 owned by the user may periodically share status information with other user terminals 120 . Here, the state information may include an update time of the offset recommendation model, a computing environment (eg, CPU usage rate, memory usage rate, etc.). When the computing environment is not good, for example, when the load of the CPU is greater than or equal to a threshold, the user terminal 120 transmits the content usage history to the other user terminal 120 having a good computing environment to offset from the other user terminal 120 . Make it usable to update the recommendation model. In addition, the user terminal 120, before starting the update of the offset recommendation model, by notifying other user terminals 120 that the update is scheduled, the offset until the updated offset recommendation model is shared with other user terminals 120 The update of the recommendation model can be stopped.

According to the embodiment of the present invention as described above, the content service server 110 models the integrated recommendation model for all users and then derives it from the offset content for each user based on the content usage history for each user. A recommendation model is generated and transmitted to each user terminal 120, and each user terminal 120 recommends the content through the offset recommendation model, thereby distributing the computing resources required to recommend the content, quickly and at low cost. We can recommend content that suits you.

In addition, in the present invention, since the offset recommendation model recommends content suitable for the characteristics of the user within the range of the content recommendation list recommended by the integrated recommendation model in each user terminal 120, the offset recommendation model is lighter than the integrated recommendation model. Therefore, compared to the content service server 110, the user terminal 120, which has a poor computing environment, does not perform many calculations, so it is efficient to use resources and saves time required for content recommendation.

In addition, in the present invention, when a user uses a plurality of user terminals 120, the offset recommendation model updated in a specific user terminal can be shared with other user terminals so that the same content service can be consistently provided in a plurality of user terminals. Make it available according to your characteristics.

While this specification contains many features, such features should not be construed as limiting the scope of the invention or the claims. Also, features described in individual embodiments herein may be implemented in combination in a single embodiment. Conversely, various features described herein in a single embodiment may be implemented in various embodiments individually, or may be implemented in appropriate combination.

Although acts have been described in the drawings in a specific order, it should not be understood that the acts are performed in the specific order as shown, or that all of the described acts are performed in a continuous order, or to obtain a desired result. . Multitasking and parallel processing can be advantageous in certain circumstances. In addition, it should be understood that the division of various system components in the above-described embodiments does not require such division in all embodiments. The program components and systems described above may generally be implemented as a package in a single software product or multiple software products.

The method of the present invention as described above may be implemented as a program and stored in a computer-readable form in a recording medium (CD-ROM, RAM, ROM, floppy disk, hard disk, magneto-optical disk, etc.). Since this process can be easily performed by a person skilled in the art to which the present invention pertains, it will not be described in detail any more.

The present invention described above, for those of ordinary skill in the art to which the present invention pertains, various substitutions, modifications and changes are possible without departing from the technical spirit of the present invention. It is not limited by the drawing.

110: content service server
120: user terminal
130: communication network
210: member registration processing unit
220: service provider
230: integrated recommendation model generation unit
240: offset recommendation model generation unit
310: communication department
320: input unit
330: service department
340: recommendation
350: common

Claims (21)

A user terminal for recommending content to a user, comprising:
a communication unit communicating with the content service server and receiving the user's offset recommendation model derived from the integrated recommendation model learned from the content service server;
an input unit for receiving an input for using content from the user;
a service unit for requesting, receiving, and outputting contents to the contents service server through the communication unit based on the input; and
and a recommendation unit for updating the offset recommendation model based on the content usage history of the user and recommending content to the user through the updated offset recommendation model. The method of claim 1,
The communication unit receives a content recommendation list recommended by the integrated recommendation model from the content service server,
The recommendation unit is configured to generate a content recommendation list unique to the user based on the content recommendation list by using the offset recommendation model. 3. The method of claim 2,
The recommendation is
and rearranging the received content recommendation list using the offset recommendation model to generate a content recommendation list unique to the user. The method of claim 1,
The offset recommendation model received by the communication unit is,
The user terminal, characterized in that the content service server is generated by deriving from the integrated recommendation model based on a difference between the content recommendation result of the integrated recommendation model and the user's content use history. The method of claim 1,
The content usage history is
A user terminal comprising at least one of a user's speech, viewing, purchase, or star rating history. 6. The method of claim 5,
As a user's viewing history, a vector value for each content is calculated by Equation 1 or Equation 2 below and used for learning.
(Equation 1)
Viewing history vector value for each content = 1+log (total watch time/running time of that content)
(Equation 2)
Viewing history vector value for each content = 1+log(total watch time/running time of that content)+log(α/number of views). where α=1 or determined proportional to the running time of the content. 6. The method of claim 5,
A user terminal, characterized in that, as a user's purchase history, a standardized vector value is calculated and used for learning by using the user's purchase amount for each content and the average amount and variance of all users who purchased each content. 6. The method of claim 5,
As a user's star point history, a standardized vector value is calculated and used for learning by using the star point given by the user for each content and the average and variance of the star point given by all users for each content. The method of claim 1,
A user terminal further comprising a sharing unit for transmitting and sharing the updated offset recommendation model to another user terminal. 10. The method of claim 9,
The sharing unit, based on the computing environment of the user terminal, the user terminal, characterized in that the user's content use history by transmitting and sharing to the other user terminal. 10. The method of claim 9,
The sharing unit, before updating the offset recommendation model, the user terminal, characterized in that the update is scheduled to be notified to the other user terminal to stop the update of the offset recommendation model of the other user terminal. A method of recommending content to a user in a user terminal, comprising:
receiving the user's offset recommendation model derived from the integrated recommendation model learned from the content service server from the content service server;
receiving an input for using content from the user;
requesting, receiving, and outputting content from the content service server based on the input; and
and updating the offset recommendation model based on the content usage history of the user and recommending content to the user through the updated offset recommendation model. 13. The method of claim 12,
Further comprising the step of receiving a content recommendation list recommended by the integrated recommendation model from the content service server,
The recommended steps are:
and generating a content recommendation list unique to the user based on the content recommendation list using the offset recommendation model. 14. The method of claim 13,
The recommended steps are:
and reordering the received content recommendation list using the offset recommendation model to generate a content recommendation list unique to the user. 13. The method of claim 12,
The offset recommendation model received in the receiving step is,
The method according to claim 1, wherein the content service server is derived and generated from the integrated recommendation model based on a difference between the content recommendation result of the integrated recommendation model and the content use history of the user. 13. The method of claim 12,
The content usage history is
A method comprising at least one of a user's speech, viewing, purchase, or star rating history. 17. The method of claim 16,
As a user's viewing history, a vector value for each content is calculated by Equation 3 or Equation 4 below and used for learning.
(Equation 3)
Viewing history vector value for each content = 1+log (total watch time/running time of that content)
(Equation 4)
Viewing history vector value for each content = 1+log(total watch time/running time of that content)+log(α/number of views). where α=1 or determined proportional to the running time of the content. 17. The method of claim 16,
A method, characterized in that, as a user's purchase history, a standardized vector value is calculated and used for learning by using the user's purchase amount for each content and the average amount and variance of all users who purchased each content. 17. The method of claim 16,
As a user's star rating history, a standardized vector value is calculated and used for learning by using the star point given by the user for each content and the average and variance of the star point given by all users for each content. 13. The method of claim 12,
The method further comprising the step of transmitting and sharing the updated offset recommendation model to another user terminal. 21. The method of claim 20,
Before updating the offset recommendation model, the method further comprising the step of notifying the other user terminal that the update is scheduled to stop, and stopping the update of the offset recommendation model of the other user terminal.

Images