KR20210102727A - Method and apparatus for recommending content based on content viewing history of user - Google Patents

Abstract

In order to recommend content to a user, a list of one or more contents viewed by the user, information of the user, and meta information of each of the contents are received as inputs. A learning matrix is generated by removing noise values in the generated matrix based on the inputs. A first characteristic representing explicit information is generated. A second characteristic representing implicit information is generated. A third characteristic indicating a characteristic of the continuity of the contents is generated. Content to the user is recommended based on the first characteristic, the second characteristic, and the third characteristic.

Description

Method and device for recommending content based on a user's content viewing list

The following embodiments relate to a technology for providing content to a user, and more specifically, to a technology for recommending content to a user based on a content list viewed by the user.

Currently, services for providing content to user terminals through wired and wireless Internet are common, and as such, the types of content provided to user terminals include image content, music content, video content, game content, real-time information content, etc. and the variety of its types is gradually increasing.

However, in order to use such content, the user has to pay a lot of time and communication cost to access the content providing server and search for the content he or she wants. Conventionally, various methods for selecting and recommending content satisfactory to a user have been proposed. For example, there is a simple recommendation method using content sales volume, purchaser's recommendation information, and the like. However, in the case of a simple recommendation method based on sales volume, etc., there is a problem in that recommendation accuracy is lowered because a user's personal taste cannot be reflected. As another example, there is a customized recommendation method using preference genre information or preference category information input by a user, or a purchase history of the user. In the case of such a customized recommendation method, since only the user's preference information is considered, there is a limit in recommending the optimal content customized to the user's preference in consideration of all continuously changing and added contents.

An embodiment may provide a method and apparatus for recommending content to a user.

An embodiment may provide a method and apparatus for recommending content to a user in consideration of various characteristics of the user.

According to an aspect, receiving a list of one or more contents viewed by a user, information of the user, and meta information of each of the contents as inputs, generating a matrix based on the input, in the matrix generating a learning matrix by removing noise values, generating a first feature representative of explicit information based on the input and the learning matrix, and generating a second feature representative of implicit information based on the learning matrix. generating a third characteristic indicating a characteristic of the continuity of the contents based on a viewing order of the contents; and providing contents to the user based on the first characteristic, the second characteristic, and the third characteristic. including the recommended steps.

The generating of the matrix based on the input may include: determining a list of one or more contents viewed by the user, the user's information, and a numeric value corresponding to each of the meta information of each of the contents; setting a numeric value to the component in the matrix.

The generating of the learning matrix by removing the noise value in the matrix includes predicting a numeric value for a target component for which a numeric value is not set among components in the matrix, and applying the predicted numeric value to the target component. It may include removing the noise value by setting.

The generating of the learning matrix may include generating the learning matrix using a dense embedding layer.

Generating a first feature representing explicit information based on the input and the learning matrix comprises adding or inner producting the information of the input and components in the learning matrix to the first feature. It may include the step of generating

Generating a first feature representative of explicit information based on the input and the learning matrix may include generating a first feature representative of explicit information based on the input and the learning matrix using an FM layer. may include

The generating of the second feature representing implicit information based on the learning matrix may include generating the second feature by inputting the learning matrix into a deep neural network (DNN) layer.

The generating of the third characteristic indicating the continuity of the contents based on the viewing order of the contents may include sequentially inputting the contents to an evolving layer based on a Gated Recurrent Unit (GRU) method. thereby creating a third characteristic.

The contents may be video on demand (VOD).

According to another aspect, the server includes a memory in which a program for recommending content to a user is recorded, and a processor for executing the program, wherein the program includes a list of one or more content viewed by the user, the user Receiving information of and meta-information of each of the contents as input, generating a matrix based on the input, generating a learning matrix by removing noise values in the matrix, in the input and the learning matrix generating a first feature representing explicit information based on the first feature, generating a second feature representing implicit information based on the learning matrix, representing a feature for the continuity of the content based on the viewing order of the content generating a third characteristic; and recommending content to the user based on the first characteristic, the second characteristic, and the third characteristic.

The generating of the matrix based on the input may include: determining a list of one or more contents viewed by the user, the user's information, and a numeric value corresponding to each of the meta information of each of the contents; setting a numeric value to the component in the matrix.

The generating of the learning matrix by removing the noise value in the matrix includes predicting a numeric value for a target component for which a numeric value is not set among components in the matrix, and applying the predicted numeric value to the target component. It may include removing the noise value by setting.

Generating a first feature representing explicit information based on the input and the learning matrix comprises adding or inner producting the information of the input and components in the learning matrix to the first feature. It may include the step of generating

The generating of the second feature representing implicit information based on the learning matrix may include generating the second feature by inputting the learning matrix into a deep neural network (DNN) layer.

The generating of the third characteristic indicating the continuity of the contents based on the viewing order of the contents may include sequentially inputting the contents to an evolving layer based on a Gated Recurrent Unit (GRU) method. thereby creating a third characteristic.

The contents may be video on demand (VOD).

A method and apparatus for recommending content to a user may be provided.

A method and apparatus for recommending content to a user in consideration of various characteristics of the user may be provided.

1 is a content providing system according to an example.
2 is a configuration diagram of a server that provides content according to an embodiment.
3 is a flowchart of a method of recommending content according to an exemplary embodiment.
4 illustrates layers of an algorithm for recommending content according to an embodiment.
5 is a flowchart of a method of generating a matrix based on an input according to an example.
6 illustrates a matrix generated based on an input according to an example.
7 is a flowchart of a method of generating a learning matrix by removing noise values in the matrix according to an example.
8 illustrates a learning matrix according to an example.
9 illustrates a method for generating a first feature representing explicit information based on an input and a learning matrix according to an example.
10 illustrates a method of generating a second feature representing implicit information based on a learning matrix according to an example.
11 illustrates a method of generating a third characteristic indicating a characteristic of continuity of contents based on a viewing order of contents according to an example.

Hereinafter, embodiments will be described in detail with reference to the accompanying drawings. However, the scope of the patent application is not limited or limited by these examples. Like reference numerals in each figure indicate like elements.

Various modifications may be made to the embodiments described below. It should be understood that the embodiments described below are not intended to limit the embodiments, and include all modifications, equivalents or substitutes thereto.

Terms used in the examples are only used to describe specific examples, and are not intended to limit the examples. The singular expression includes the plural expression unless the context clearly dictates otherwise. In the present specification, terms such as “comprise” or “have” are intended to designate that a feature, number, step, operation, component, part, or combination thereof described in the specification exists, but one or more other features It should be understood that this does not preclude the existence or addition of numbers, steps, operations, components, parts, or combinations thereof.

Unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art to which the embodiment belongs. Terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with the meaning in the context of the related art, and should not be interpreted in an ideal or excessively formal meaning unless explicitly defined in the present application. does not

In addition, in the description with reference to the accompanying drawings, the same components are given the same reference numerals regardless of the reference numerals, and the overlapping description thereof will be omitted. In the description of the embodiment, if it is determined that a detailed description of a related known technology may unnecessarily obscure the gist of the embodiment, the detailed description thereof will be omitted.

1 is a content providing system according to an example.

According to one aspect, the content providing system includes a user terminal 120 and a server 110 that provides content to the user terminal 120 . For example, the server 110 provides a list of contents to the user terminal 120 , and the user of the user terminal 120 may select any one of the list. The content may be video on demand (VOD), and the server 110 may provide the VOD to the user terminal 120 in a streaming manner.

The server 110 may determine the target content that the user likes, and recommend the target content to the user as the recommended content. In order to determine the target content among the plurality of contents, various factors may be considered. For example, user information including the user's age and gender, and the user's viewing history may be considered.

A trainable algorithm may be used to determine the target content. Algorithms can be trained based on feedback on various inputs and outputs. For example, the feedback may include a user's rating, etc., but is not limited to the described embodiment.

Hereinafter, a method of recommending content to a user using an algorithm will be described in detail with reference to FIGS. 2 to 11 .

2 is a configuration diagram of a server that provides content according to an embodiment.

The server 200 includes a communication unit 210 , a processor 220 , and a memory 230 . For example, the server 200 may be the server 110 with reference to FIG. 1 .

The communication unit 210 is connected to the processor 220 and the memory 230 to transmit and receive data. The communication unit 210 may be connected to another external device to transmit/receive data. Hereinafter, the expression "transmitting and receiving "A" may indicate transmitting and receiving "information or data representing A".

The communication unit 210 may be implemented as circuitry in the server 200 . For example, the communication unit 210 may include an internal bus and an external bus. As another example, the communication unit 210 may be an element that connects the server 200 and an external device. The communication unit 210 may be an interface. The communication unit 210 may receive data from an external device and transmit the data to the processor 220 and the memory 230 .

The processor 220 processes data received by the communication unit 210 and data stored in the memory 230 . A “processor” may be a data processing device implemented in hardware having circuitry having a physical structure for performing desired operations. For example, desired operations may include code or instructions included in a program. For example, a data processing device implemented as hardware includes a microprocessor, a central processing unit, a processor core, a multi-core processor, and a multiprocessor. , an Application-Specific Integrated Circuit (ASIC), and a Field Programmable Gate Array (FPGA).

Processor 220 executes computer readable code (eg, software) stored in memory (eg, memory 230 ) and instructions issued by processor 220 .

The memory 230 stores data received by the communication unit 210 and data processed by the processor 220 . For example, the memory 230 may store a program (or an application, software). The stored program may be a set of syntaxes coded to recommend content and executable by the processor 220 .

According to one aspect, memory 230 may include one or more of volatile memory, non-volatile memory and random access memory (RAM), flash memory, hard disk drive, and optical disk drive.

The memory 230 stores an instruction set (eg, software) for operating the server 200 . The instruction set for operating the server 200 is executed by the processor 220 .

The communication unit 210 , the processor 220 , and the memory 230 will be described in detail below with reference to FIGS. 3 to 11 .

3 is a flowchart of a method of recommending content according to an exemplary embodiment.

The steps 310 to 370 below are performed by the server 200 described above with reference to FIG. 2 . The server 200 may include an algorithm for recommending content, and steps 310 to 370 may be performed using the algorithm. Steps 310 - 370 may be performed on the target user. For example, a target user may be identified by a user identifier (ID) or a user's session.

According to one aspect, the algorithm may be the algorithm 400 of FIG. 4 , and the algorithm 400 may include, for example, 7 layers. Each layer of the algorithm 400 is described along with a description of steps 310-370.

In step 310 , the server 200 receives a list of one or more contents viewed by the user, user information, and meta information of each of the contents as inputs. For example, the user's information may include the user's age, content preference genre, gender, subscription product, and the like, and is not limited to the described embodiment. For example, the meta information of the content may be information mapped or tagged to the content. The meta information of the content may be a keyword indicating content characteristics of the content.

According to one aspect, step 310 may be performed in the input layer 410 of the algorithm 400 . For example, the input layer 410 may include a plurality of nodes, and a corresponding input may be set in the node. If there is no information corresponding to the target node, the input may not be received at the target node.

In step 320, the server 200 generates a matrix based on the input. For example, the matrix may be generated in the sparse embedding layer 410 of the algorithm 400 . For example, each piece of information as input may be converted into a numeric value, and the converted numeric value may be set to a corresponding component in a matrix. Components within a matrix may represent rows and columns of the matrix. When there is no information corresponding to a specific component in the matrix, the corresponding component may be an empty space in which a numeric value is not set. A method of generating the matrix is described in detail below with reference to FIGS. 5 and 6 .

In step 330, the server 200 generates a learning matrix by removing noise values in the matrix. For example, the noise value may be a component in a matrix for which a numeric value is not set. The server 200 may remove a noise value by predicting a numeric value for the target component and setting the predicted numeric value to the target component.

According to one aspect, step 330 may be performed by a dense embedding layer 430 of the algorithm 400 .

A method of generating a learning matrix is described in detail below with reference to FIGS. 7 and 8 .

In step 340, the server 200 generates a first feature representing explicit information based on the input and the learning matrix. According to one aspect, step 340 may be performed by the FM layer 440 of the algorithm 400 . A method of generating the first feature is described in detail below with reference to FIG. 9 .

In step 350, server 200 generates a second feature representing implicit information based on the learning matrix. According to one aspect, step 350 may be performed by a deep neural network (DNN) layer 450 of the algorithm 400 . A method of generating the second feature is described in detail below with reference to FIG. 10 .

In step 360, the server 200 generates a third characteristic indicating a characteristic of the continuity of the contents based on the viewing order of the contents. According to one aspect, step 360 may be performed by an evolving layer 460 of algorithm 400 . A method of generating the third feature is described in detail below with reference to FIG. 11 .

In step 370 , the server 200 recommends content to the user based on the first characteristic, the second characteristic, and the third characteristic. According to one aspect, step 370 may be performed by the output layer 470 of the algorithm 400 . Content to be recommended to a user may be determined using an activation function of the output layer 470 . For example, the activation function may be a sigmoid function.

4 illustrates layers of an algorithm for recommending content according to an embodiment.

According to an aspect, the content recommendation algorithm 400 includes the input layer 410, the sparse embedding layer 420, the dense embedding layer 430, the FM layer 440, and the DNN layer described above with reference to FIG. 3 . 450 , an evolving layer 460 , and an output layer 470 . Each of the layers and the nodes within the layers in the algorithm 400 may be customized and pre-trained to recommend content to a user.

Algorithm 400 is described in detail below with reference to FIGS. 5-11 .

5 is a flowchart of a method of generating a matrix based on an input according to an example.

According to one aspect, the step 320 described above with reference to FIG. 3 may include the following steps 510 and 520 .

In step 510 , the server 200 determines a numeric value corresponding to a list of contents, user information, and meta information of the contents. A numeric value corresponding to the information may be determined for processing the information appearing as text. For example, the range of numeric values can be from 0 to 1.

In step 520, the server 200 sets the determined numerical value for the target information to the target element in the matrix. When information corresponding to the target component is not received, a numeric value may not be set for the target component. A component for which a numeric value is not set may be noise.

6 illustrates a matrix generated based on an input according to an example.

For example, the matrix 600 in the sparse embedding layer 420 may include a first component 610 , a second component 620 , and a third component 630 . When the first information 640 and the second information 650 are received as inputs, the numerical values 641 and 642 for the first information 640 are set in the second component 620, and the second information ( Numeric values 651 and 652 for 650 may be set in third component 630 . For example, the numeric values 641 and 642 may be represented as vectors.

When information corresponding to the third component 630 is not received, the third component 630 may be noise.

7 is a flowchart of a method of generating a learning matrix by removing noise values in the matrix according to an example.

According to one aspect, step 330 described above with reference to FIG. 3 may include steps 710 and 720 below.

In step 710 , the server 200 predicts a numeric value for a target component to which a numeric value is not set among components in the matrix. For example, a numeric value may be predicted based on surrounding components of the target component.

In step 720, the server 200 removes the noise value by setting the predicted numeric value to the target component. A learning matrix may be generated by removing the noise value of the matrix.

8 illustrates a learning matrix according to an example.

Since information corresponding to the first component 610 of the matrix 600 described above with reference to FIG. 6 is not received, the first component 610 may be noise. Numeric values 801 and 802 for the first component 610 may be predicted. For example, numeric values 801 and 802 may be predicted based on numeric values 641 , 642 , 651 and 652 .

A learning matrix may be created by setting the numerical values 801 and 802 in the first component 610 .

9 illustrates a method for generating a first feature representing explicit information based on an input and a learning matrix according to an example.

According to one aspect, step 340 described above with reference to FIG. 3 may be performed by the FM layer 440 of the algorithm 400 .

For example, the FM layer 440 may include a first node 910 , a second node 920 , and a third node 930 . The result values of the first node 910 , the second node 920 , and the third node 930 may be a first characteristic indicating explicit information.

The first node 910 may add the first information 640 and the second information 650 . The second node 920 and the third node 930 may inner product the components 610 , 620 and 630 in the learning matrix.

10 illustrates a method of generating a second feature representing implicit information based on a learning matrix according to an example.

According to one aspect, the step 350 described above with reference to FIG. 3 may be performed by the DNN layer 450 of the algorithm 400 . The DNN layer 450 may include a plurality of layers 1010 and 1020 . The DNN layer 450 may be a hidden layer.

Components 610 , 620 and 630 in the learning matrix may be input to the DNN layer 450 , and a second feature 1030 may be generated as an output.

11 illustrates a method of generating a third characteristic indicating a characteristic of continuity of contents based on a viewing order of contents according to an example.

According to one aspect, step 360 described above with reference to FIG. 3 may be performed by the evolving layer 450 of the algorithm 400 . The evolving layer 450 may be pre-trained based on the viewing order of the contents viewed by the user. Contents may be sequentially input to the evolving layer 450 based on a Gated Recurrent Unit (GRU) scheme. h may indicate a third characteristic for the user, and h _o to h _t indicate a third characteristic that is updated whenever information (e ₁ to e _{t ) on new content is input.} For example, the information e about the content may include meta information of the content.

The device described above may be implemented as a hardware component, a software component, and/or a combination of the hardware component and the software component. For example, devices and components described in the embodiments may include, for example, a processor, a controller, an arithmetic logic unit (ALU), a digital signal processor, a microcomputer, a field programmable array (FPA), It may be implemented using one or more general purpose or special purpose computers, such as a programmable logic unit (PLU), microprocessor, or any other device capable of executing and responding to instructions. The processing device may execute an operating system (OS) and one or more software applications running on the operating system. The processing device may also access, store, manipulate, process, and generate data in response to execution of the software. For convenience of understanding, although one processing device is sometimes described as being used, one of ordinary skill in the art will recognize that the processing device includes a plurality of processing elements and/or a plurality of types of processing elements. It can be seen that can include For example, the processing device may include a plurality of processors or one processor and one controller. Other processing configurations are also possible, such as parallel processors.

The method according to the embodiment may be implemented in the form of program instructions that can be executed through various computer means and recorded in a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, etc. alone or in combination. The program instructions recorded on the medium may be specially designed and configured for the embodiment, or may be known and available to those skilled in the art of computer software. Examples of the computer-readable recording medium include magnetic media such as hard disks, floppy disks and magnetic tapes, optical media such as CD-ROMs and DVDs, and magnetic such as floppy disks. - includes magneto-optical media, and hardware devices specially configured to store and carry out program instructions, such as ROM, RAM, flash memory, and the like. Examples of program instructions include not only machine language codes such as those generated by a compiler, but also high-level language codes that can be executed by a computer using an interpreter or the like. The hardware devices described above may be configured to operate as one or more software modules to perform the operations of the embodiments, and vice versa.

As described above, although the embodiments have been described with reference to the limited embodiments and drawings, various modifications and variations are possible from the above description by those skilled in the art. For example, the described techniques are performed in a different order than the described method, and/or the described components of the system, structure, apparatus, circuit, etc. are combined or combined in a different form than the described method, or other components Or substituted or substituted by equivalents may achieve an appropriate result.

Claims (17)

receiving a list of one or more contents viewed by a user, information of the user, and meta information of each of the contents as inputs;
generating a matrix based on the input;
generating a learning matrix by removing noise values in the matrix;
generating a first feature representing explicit information based on the input and the learning matrix;
generating a second feature representing implicit information based on the learning matrix;
generating a third characteristic indicating a characteristic of the continuity of the contents based on the viewing order of the contents; and
recommending content to the user based on the first characteristic, the second characteristic, and the third characteristic;
containing,
How to recommend content.
According to claim 1,
The step of generating a matrix based on the input comprises:
determining a list of one or more contents viewed by a user, a numeric value corresponding to each of the user's information and meta information of each of the contents; and
setting the determined numerical value to a component in the matrix;
containing,
How to recommend content.
3. The method of claim 2,
The step of generating a learning matrix by removing noise values in the matrix comprises:
predicting a numeric value for a target component to which a numeric value is not set among components in the matrix; and
removing the noise value by setting the predicted numeric value to the target component;
containing,
How to recommend content.
4. The method of claim 3,
The step of generating the learning matrix comprises:
Generating the learning matrix using a dense embedding layer
containing,
How to recommend content.
According to claim 1,
generating a first feature representing explicit information based on the input and the learning matrix comprises:
generating the first feature by adding or inner producting the information of the input and components in the learning matrix;
containing,
How to recommend content.
6. The method of claim 5,
generating a first feature representing explicit information based on the input and the learning matrix comprises:
generating a first feature representing explicit information based on the input and the learning matrix using an FM layer;
containing,
How to recommend content.
According to claim 1,
Generating a second feature representing implicit information based on the learning matrix comprises:
generating the second feature by inputting the learning matrix into a deep neural network (DNN) layer;
containing,
How to recommend content.
According to claim 1,
The step of generating a third characteristic indicating a characteristic of the continuity of the contents based on the viewing order of the contents,
Generating the third feature by sequentially inputting the contents to an evolving layer based on a Gated Recurrent Unit (GRU) scheme
containing,
How to recommend content.
According to claim 1,
The contents are video on demand (VOD),
How to recommend content.
A computer-readable recording medium containing a program for performing the method of any one of claims 1 to 9.
server,
a memory in which a program for recommending content to a user is recorded; and
a processor that executes the program
including,
The program is
receiving a list of one or more contents viewed by a user, information of the user, and meta information of each of the contents as inputs;
generating a matrix based on the input;
generating a learning matrix by removing noise values in the matrix;
generating a first feature representing explicit information based on the input and the learning matrix;
generating a second feature representing implicit information based on the learning matrix;
generating a third characteristic indicating a characteristic of the continuity of the contents based on the viewing order of the contents; and
recommending content to the user based on the first characteristic, the second characteristic, and the third characteristic;
to do,
server.
12. The method of claim 11,
The step of generating a matrix based on the input comprises:
determining a list of one or more contents viewed by a user, a numeric value corresponding to each of the user's information and meta information of each of the contents; and
setting the determined numerical value to a component in the matrix;
containing,
server.
13. The method of claim 12,
The step of generating a learning matrix by removing noise values in the matrix comprises:
predicting a numeric value for a target component to which a numeric value is not set among components in the matrix; and
removing the noise value by setting the predicted numeric value to the target component;
containing,
server.
12. The method of claim 11,
generating a first feature representing explicit information based on the input and the learning matrix comprises:
generating the first feature by adding or inner producting the information of the input and components in the learning matrix;
containing,
server.
12. The method of claim 11,
Generating a second feature representing implicit information based on the learning matrix comprises:
generating the second feature by inputting the learning matrix into a deep neural network (DNN) layer;
containing,
server.
12. The method of claim 11,
The step of generating a third characteristic indicating a characteristic of the continuity of the contents based on the viewing order of the contents,
Generating the third feature by sequentially inputting the contents to an evolving layer based on a Gated Recurrent Unit (GRU) scheme
containing,
server.
12. The method of claim 11,
The contents are video on demand (VOD),
server.

Images