KR102342522B1 - Recommendation apparatus and operating merhod thereof - Google Patents

Abstract

A recommendation device is disclosed. An embodiment generates an optimal keyword selection prediction matrix for the users and the keywords by iteratively learning a first feature vector for users and a second feature vector for keywords, and selects the generated optimal keyword Determine preferred keywords for one or more of the users based on a prediction matrix, and provide the determined preferred keywords to one or more of the users.

Description

RECOMMENDATION APPARATUS AND OPERATING MERHOD THEREOF

The following embodiments relate to a recommendation device.

In the past, content was only recommended to users, but various keywords were not recommended.

As a related prior art, there is Korean Patent Registration No. 10-2004145 (Title of Invention: Method and Apparatus for Content Recommendation, Applicant: Korea Institute of Science and Technology Information). The content recommendation method disclosed in the corresponding registered patent publication includes the steps of obtaining reference content, obtaining information on the expanded co-author by using co-author information of the reference content, and based on the obtained information on the expanded co-author, a plurality of Steps to get the recommended candidate content of . calculating a similarity between each recommended candidate content included in the plurality of recommended candidate content and the reference content, selecting some of the plurality of recommended candidate content based on the similarity, and using the selected content as recommended content outputting it.

According to one aspect, an operating method of a recommendation apparatus includes generating an optimal keyword selection prediction matrix for the users and the keywords by iteratively learning a first feature vector for users and a second feature vector for keywords ; determining preferred keywords for one or more of the users based on the generated optimal keyword selection prediction matrix; and providing the determined preferred keyword to one or more of the users.

The generating may include transforming the information on the users into a user matrix and transforming the information on the keywords into a keyword matrix; and generating the first feature vector and the second feature vector based on the transformed user matrix and the transformed keyword matrix.

A matrix in which information on content related to each of the keywords is converted may be added to the keyword matrix.

Each of the user matrix and the keyword matrix may correspond to a diagonal matrix.

The generating may include: performing matrix multiplication on the first feature vector and the second feature vector; generating a difference matrix corresponding to a difference between a selection history matrix for the keyword selection history of the users and a result of the matrix multiplication; generating an error matrix based on a preference matrix related to the user's preferences for the keywords and the generated difference matrix; and learning the first feature vector and the second feature vector based on the generated error matrix.

The generated optimal keyword selection prediction matrix may include a prediction probability value for each of the users to select each of the keywords.

The determining may include determining, as the preferred keyword of the individual user, a keyword corresponding to the highest predicted probability value among the predicted probability values for the individual user.

The method of operating the recommendation device may further include recommending content related to the determined preferred keyword to one or more of the users.

The recommendation apparatus according to one aspect generates an optimal keyword selection prediction matrix for the users and the keywords by iteratively learning a first feature vector for users and a second feature vector for keywords, and the generated and a processor that determines preferred keywords for one or more of the users based on an optimal keyword selection prediction matrix, and provides the determined preferred keywords to one or more of the users.

The processor converts the information about the users into a user matrix, converts the information on the keywords into a keyword matrix, and the first feature vector and the second based on the transformed user matrix and the transformed keyword matrix. Two feature vectors can be created.

A matrix in which information on content related to each of the keywords is converted may be added to the keyword matrix.

Each of the user matrix and the keyword matrix may correspond to a diagonal matrix.

The processor performs matrix multiplication on the first feature vector and the second feature vector, and generates a difference matrix corresponding to a difference between a selection history matrix for the keyword selection history of the users and a result of the matrix multiplication, , generate an error matrix based on the generated difference matrix and a preference matrix related to the users' preferences for the keywords, and learn the first feature vector and the second feature vector based on the generated error matrix can do it

The generated optimal keyword selection prediction matrix may include a prediction probability value for each of the users to select each of the keywords.

The processor may determine a keyword corresponding to the largest predicted probability value among the predicted probability values for the individual user as the preferred keyword of the individual user.

The processor may recommend content related to the determined preferred keyword to one or more of the users.

Embodiments may recommend keywords to users, and content related to the keyword may be recommended to users together with the keyword, so that various recommendations may be provided to users.

In addition, embodiments may recommend customized preferred keywords based on user history to the user, thereby improving the user's VOD viewing and purchasing rate, and increasing customer satisfaction.

1 is a diagram for explaining a recommendation system according to an embodiment.
2 to 7 are diagrams for explaining an operation of a recommendation device in a recommendation system according to an exemplary embodiment.
8 is a flowchart illustrating a method of operating a recommendation apparatus according to an exemplary embodiment.
9 is a block diagram illustrating a recommendation apparatus according to an exemplary embodiment.

Hereinafter, embodiments will be described in detail with reference to the accompanying drawings. However, since various changes may be made to the embodiments, the scope of the patent application is not limited or limited by these embodiments. It should be understood that all modifications, equivalents and substitutes for the embodiments are included in the scope of the rights.

The terms used in the examples are used for the purpose of description only, and should not be construed as limiting. The singular expression includes the plural expression unless the context clearly dictates otherwise. In this specification, terms such as "comprise" or "have" are intended to designate that a feature, number, step, operation, component, part, or a 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 describing 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 diagram for explaining a recommendation system according to an embodiment.

Referring to FIG. 1 , the recommendation system 100 includes a recommendation device 110 and content reproducing devices 120-1 to 120-n.

The recommendation device 110 may be expressed differently as a recommendation server. As will be described later, the recommendation device 110 may recommend a user-customized preference keyword to the user based on the factorization machine. Also, the recommendation device 110 may recommend content related to a preferred keyword to the user. Accordingly, the recommendation device 110 may be expressed differently as a keyword/content recommendation device.

The recommendation system 100 may support, for example, an Internet Protocol TV (IPTV) system. One or more of the content playback devices 120 - 1 to 120 - n may include a set-top box and a display. The recommendation device 110 may recommend or transmit keywords and/or content (eg, real-time content and/or on-demand content) for a user to the set-top box. Users can search for content through recommended keywords.

Also, the recommendation system 100 may support a mobile content system. One or more of the content playback devices 120-1 to 120-n may represent a mobile terminal (eg, a smartphone, a tablet, etc.), and the recommendation device 110 may provide keywords and/or content to the mobile terminal. You can recommend or send.

Hereinafter, an operation of the recommendation device 110 will be described with reference to FIGS. 2 to 6 .

2 to 6 are diagrams for explaining an operation of a recommendation apparatus in a recommendation system according to an exemplary embodiment.

Referring to FIG. 2 , the recommendation apparatus 110 may convert information about users (eg, IDs of users) into a user matrix 210 . The user matrix 210 may correspond to, for example, a diagonal matrix.

In FIG. 2 , a matrix 210 - 1 represents an example of a user matrix 210 . Each row and column of the matrix 210-1 represents users 1 to n, and the diagonal components of the matrix 210-1 are 1.

The recommendation apparatus 110 may convert information on keywords (eg, an ID of each keyword) into the keyword matrix 220 . Each of the keywords may correspond to a word used to search for content. The keyword matrix 220 may correspond to, for example, a diagonal matrix.

In FIG. 2 , a matrix 220 - 1 represents an example of a keyword matrix 220 . Each row and column of the matrix 220-1 represents keywords 1 to m, and the diagonal elements of the matrix 220-1 are 1.

According to an embodiment, the recommendation apparatus 110 may generate the matrix 230 by transforming information on content related to each of the keywords. For example, when the keywords used to search for content 2 are keywords 1 and 2, each of the components (1,2) and (2,2) of the matrix 230 may be 1. In other words, when content 2 is included in each of keywords 1 and 2, each of the (1,2) and (2,2) components of the matrix 230 may be 1. In this way, the recommendation apparatus 110 may generate the matrix 230 by transforming information on content related to each of the keywords. An example 230-1 of the matrix 230 is shown in FIG.

As in the example shown in FIG. 2 , the recommendation apparatus 110 may add or connect the matrix 230 to the keyword matrix 220 .

Referring to FIG. 3 , the recommendation device 110 generates a first feature vector 310 for users and a second feature vector 320 for keywords based on a user matrix 210 and a keyword matrix 220 . can create In this case, as in the example shown in FIG. 3 , a third feature vector 330 for content may be added or connected to the second feature vector 320 .

The user matrix 210 and the keyword matrix 220 are mostly composed of 0, and the recommendation device 110 can efficiently convert the sparse matrices 210 and 220 composed of mostly 0 into a learnable structure. . For example, the recommendation device 110 may use a factorization machine.

In FIG. 3 , the matrix 310-1 represents an example of the first feature vector 310, the matrix 320-1 represents an example of the second feature vector 320, and the matrix 330-1 represents the third An example of the feature vector 330 is shown. For convenience of description, each component of the matrices 310-1, 320-1, and 330-1 is omitted.

(410)을 생성할 수 있다. 키워드 선택 예측 행렬

(410)은 사용자들 각각이 각 키워드를 선택할 예측 확률값을 포함할 수 있다. Referring to FIG. 4 , the recommendation apparatus 110 performs matrix multiplication on the first feature vector 310 and the second feature vector 320 to obtain a keyword selection prediction matrix.

410 can be created. Keyword Selection Prediction Matrix

410 may include a predicted probability value for each of the users to select each keyword.

(410)의 일례를 나타낸다. 행렬(410-1)에서 (1,1) 성분은 사용자 1이 키워드 1을 선택(또는 선호)할 예측 확률값을 나타내고, (1,m) 성분은 사용자 1이 키워드 m을 선택(또는 선호)할 예측 확률값을 나타낸다. 또한, 행렬(410-1)에서, (n,1) 성분은 사용자 n이 키워드 1을 선택(또는 선호)할 예측 확률값을 나타내고, (n,m) 성분은 사용자 n이 키워드 m을 선택(또는 선호)할 예측 확률값을 나타낸다.In FIG. 4 , a matrix 410 - 1 is a keyword selection prediction matrix

An example of (410) is shown. In the matrix 410-1, the (1,1) component represents a predicted probability value that user 1 will select (or prefer) keyword 1, and the (1,m) component indicates that user 1 selects (or prefers) keyword m. Indicates the predicted probability value. In addition, in the matrix 410-1, the (n,1) component represents a predicted probability value that the user n will select (or prefer) keyword 1, and the (n,m) component indicates that the user n selects (or prefers) the keyword m Preferred) to indicate the predicted probability value.

Hereinafter, learning of the first feature vector 310 and the second feature vector 320 will be described with reference to FIG. 5 .

Referring to FIG. 5 , the recommendation apparatus 110 may generate a matrix 510 by identifying whether each of the users has selected each keyword.

Rows of matrix 510 may represent users and columns may represent keywords. Depending on the implementation, rows of matrix 510 may represent keywords and columns may represent users.

Each of the elements in matrix 510 may have a first value (eg, 1) or a second value (eg, 0). As an example, the (3,2) component of the matrix 510 may have 1, and the (3,3) component may have 0. A component of (3,2) of 1 may indicate that the user 3 has selected the keyword 2, and a component of (3,3) of 0 may indicate that the user 3 does not select the keyword 3.

The recommendation device 110 may generate the matrix 520 based on each of the users' keyword preference scores and a predetermined function.

Rows of matrix 520 may represent users and columns may represent keywords. Depending on the implementation, rows of matrix 520 may represent keywords and columns may represent users.

Matrix 520 may represent, for example, a matrix for each of the users' keyword preferences. As an example, the (1,1) component of the matrix 520 may indicate user 1's preference for keyword 1 and the (1,3) component may indicate user 1's preference for keyword 3 .

(410)을 기초로 에러 행렬(530)을 계산할 수 있고, 학습 피드백 시 에러 행렬(530)을 적용하여 제1 특징 벡터(310) 및 제2 특징 벡터(320)를 학습시킬 수 있다. 이하, 도 6을 참조하면서 학습에 대해 구체적으로 설명한다.The recommendation apparatus 110 includes a matrix 510 , a matrix 520 , and a keyword selection prediction matrix

The error matrix 530 may be calculated based on 410 , and the first feature vector 310 and the second feature vector 320 may be learned by applying the error matrix 530 during learning feedback. Hereinafter, learning will be described in detail with reference to FIG. 6 .

In the example shown in FIG. 6 , it is assumed that the number of repetitions is m.

로 표현하고, 에러 행렬을

로 표현한다.The recommendation apparatus 110 may generate a keyword selection prediction matrix using the feature vectors 310 and 320 and may generate an error matrix. Since it is before training, the corresponding keyword selection prediction matrix is

expressed as , and the error matrix is

expressed as

를 활용할 수 있다. The recommendation apparatus 110 may learn the feature vectors 310 and 320 through _{training 1.} At this time, the recommendation device 110 performs the learning ₁ error matrix

can utilize

을 생성할 수 있고, 에러 행렬

을 생성할 수 있다.The recommendation apparatus 110 performs matrix multiplication on the feature vectors 310 and 320 learned through _{training 1 to obtain a keyword selection prediction matrix.}

can create an error matrix

can create

을 활용할 수 있다. 추천 장치(110)는 재학습된 특징 벡터들(310 및 330)에 매트릭스 곱셈을 수행하여 키워드 선택 예측 행렬

를 생성할 수 있고, 에러 행렬

를 생성할 수 있다. 이러한 방식으로, 추천 장치(110)는 학습을 m번 수행할 수 있고 도 6에 도시된 예와 같이, 최종 학습_m에서의 키워드 선택 예측 행렬

을 생성할 수 있다. 다시 말해, 추천 장치(110)는 비용 함수가 최적화되도록(또는 에러가 최소화되도록) 특징 벡터들(310 및 320)에 대한 학습을 반복적으로 수행할 수 있고, 학습 완료된 특징 벡터들(310 및 320)을 기초로 키워드 선택 예측 행렬

을 생성할 수 있다. The recommendation apparatus 110 may retrain the feature vectors 310 and 320 learned through _{training 1 through training 2 .} At this time, the recommendation device 110 performs an error matrix when _{learning 2 is performed.}

can utilize The recommendation apparatus 110 performs matrix multiplication on the retrained feature vectors 310 and 330 to obtain a keyword selection prediction matrix.

can create an error matrix

can create In this way, the recommendation apparatus 110 may perform learning m times and as shown in the example shown in FIG. 6 , the keyword selection prediction matrix in the _{final learning m}

can create In other words, the recommendation apparatus 110 may iteratively perform learning on the feature vectors 310 and 320 so that the cost function is optimized (or the error is minimized), and the learned feature vectors 310 and 320 . Keyword selection prediction matrix based on

can create

는 학습 완료됨으로써 생성된 것이므로, 최적 키워드 선택 예측 행렬 또는 최종 키워드 선택 예측 행렬로 달리 표현될 수 있다.Keyword Selection Prediction Matrix

Since is generated by completing learning, it may be expressed differently as an optimal keyword selection prediction matrix or a final keyword selection prediction matrix.

의 행은 사용자들을 나타낼 수 있고 열은 키워드들을 나타낼 수 있다. 구현에 따라, 키워드 선택 예측 행렬

의 행은 키워드들을 나타낼 수 있고 열은 사용자들을 나타낼 수 있다.Best Keyword Selection Prediction Matrix

Rows of may represent users and columns may represent keywords. Depending on the implementation, the keyword selection prediction matrix

Rows of may represent keywords and columns may represent users.

을 기초로 사용자들 중 하나 이상을 위한 선호 키워드를 결정할 수 있다. 일례로, 추천 장치(110)는 개별 사용자에 대한 선택 예측 확률값들 중 가장 큰 예측 확률값에 해당하는 키워드를 개별 사용자의 선호 키워드로 결정할 수 있다. 도 7에 도시된 예에서, 추천 장치(110)는 사용자 1의 각 키워드에 대한 선택 예측 확률값 p₁₁, p₁₂, …, p_1m 중에서 최대값을 찾을 수 있다. 여기서, p₁₂가 최대값이 경우, 추천 장치(110)는 사용자 1에게 키워드 2를 추천할 수 있다. 이와 함께, 추천 장치(110)는 키워드 2에 관한 콘텐츠를 사용자 1에게 추천할 수 있다.The recommendation device 110 is an optimal keyword selection prediction matrix

A preferred keyword for one or more of the users may be determined based on the . As an example, the recommendation apparatus 110 may determine a keyword corresponding to the largest prediction probability value among selection prediction probability values for an individual user as a preferred keyword of the individual user. In the example shown in FIG. 7 , the recommendation apparatus 110 selects prediction probability values p ₁₁ , p ₁₂ , ... for each keyword of user 1 . , p _1m , the maximum value can be found. Here, when p ₁₂ is the maximum value, the recommendation apparatus 110 may recommend keyword 2 to user 1 . In addition, the recommendation device 110 may recommend the content related to the keyword 2 to the user 1 .

8 is a flowchart illustrating a method of operating a recommendation apparatus according to an exemplary embodiment.

Referring to FIG. 8 , the recommendation apparatus 110 according to an embodiment repeatedly learns a first feature vector for users and a second feature vector for keywords to predict optimal keyword selection for users and keywords. A matrix is created (810).

In step 810 , the recommendation device 110 may convert information about users (eg, user IDs) into the user matrix 210 and information about keywords (eg, keyword IDs) ) can be converted into the keyword matrix 220 . The recommendation apparatus 110 may generate the first feature vector 310 and the second feature vector 320 based on the transformed user matrix and the transformed keyword matrix.

The recommendation apparatus 110 may perform matrix multiplication on the first feature vector 310 and the second feature vector 320 . The recommendation apparatus 110 may generate a difference matrix corresponding to a difference between the user's keyword selection history selection history matrix and a result of matrix multiplication. Here, the selection history matrix may correspond to the matrix 510 described with reference to FIG. 5 . The recommendation apparatus 110 may generate an error matrix based on a preference matrix related to users' preferences for keywords and the generated difference matrix. Here, the preference matrix may correspond to the matrix 520 described with reference to FIG. 5 . The recommendation apparatus 110 may learn the first feature vector 310 and the second feature vector 320 based on the generated error matrix. The recommendation apparatus 110 may iteratively learn the first feature vector 310 and the second feature vector 320 so that the error function is minimized, and perform matrix multiplication on the learned first feature vector and the second feature vector. By doing so, it is possible to generate an optimal keyword selection prediction matrix.

The recommendation apparatus 110 determines preferred keywords for one or more of the users based on the generated optimal keyword selection prediction matrix ( 820 ).

The recommendation device 110 provides the determined preferred keyword to one or more of the users ( 830 ).

Since the matters described with reference to FIGS. 1 to 7 may be applied to the matters described with reference to FIG. 8 , a detailed description thereof will be omitted.

9 is a block diagram illustrating a recommendation apparatus according to an exemplary embodiment.

Referring to FIG. 9 , the recommendation device 110 includes a memory 910 and a processor 920 .

The memory 910 stores an operation result or an operation result of the processor 920 . For example, the memory 910 may include a user matrix 210 , a keyword matrix 220 , a matrix 230 generated by converting information on content related to each of the keywords, a first feature vector 310 , and a second feature The vector 320 , the third feature vector 330 , the matrix 510 , the matrix 520 , the error matrix and keyword selection prediction matrix generated in each learning step may be stored.

The processor 920 iteratively learns a first feature vector for users and a second feature vector for keywords to generate an optimal keyword selection prediction matrix for users and keywords.

The processor 920 determines preferred keywords for one or more of the users based on the generated optimal keyword selection prediction matrix.

The processor 9210 provides the determined preferred keyword to one or more of the users.

Since the matters described with reference to FIGS. 1 to 8 may be applied to the matters described with reference to FIG. 9 , a detailed description thereof will be omitted.

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 gate array (FPGA). , a programmable logic unit (PLU), microprocessor, or any other device capable of executing and responding to instructions, may be implemented using one or more general purpose or special purpose computers. The processing device may execute an operating system (OS) and one or more software applications running on the operating system. A 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.

Software may comprise a computer program, code, instructions, or a combination of one or more thereof, which configures a processing device to operate as desired or is independently or collectively processed You can command the device. The software and/or data may be any kind of machine, component, physical device, virtual equipment, computer storage medium or apparatus, to be interpreted by or to provide instructions or data to the processing device. , or may be permanently or temporarily embody in a transmitted signal wave. The software may be distributed over networked computer systems and stored or executed in a distributed manner. Software and data may be stored in one or more computer-readable recording media.

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 execute 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 an order different from 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.

Therefore, other implementations, other embodiments, and equivalents to the claims are also within the scope of the following claims.

Claims (16)

generating an optimal keyword selection prediction matrix for the users and the keywords by iteratively learning a first feature vector for users and a second feature vector for keywords;
determining preferred keywords for one or more of the users based on the generated optimal keyword selection prediction matrix; and
providing the determined preferred keyword to one or more of the users;
including,
The generating step is
generating a first matrix indicating whether each of the users has selected each of the keywords;
generating a second matrix representing how much each of the users prefers each of the keywords; and
calculating a first keyword selection prediction matrix using the first feature vector and the second feature vector, calculating a difference between the first matrix and the calculated first keyword selection prediction matrix, the second matrix and determining an error matrix by performing matrix multiplication on the calculated difference, and learning the first feature vector and the second feature vector using the determined error matrix
including,
The generated optimal keyword selection prediction matrix includes a component (n, m), wherein the component (n, m) represents a predicted probability value at which the nth user among the users will select the mth keyword among the keywords,
How the recommended device works.
According to claim 1,
The generating step is
transforming the information on the users into a user matrix and transforming the information on the keywords into a keyword matrix; and
generating the first feature vector and the second feature vector based on the transformed user matrix and the transformed keyword matrix
containing,
How the recommended device works.
3. The method of claim 2,
A matrix in which information on content related to each of the keywords is converted is added to the keyword matrix,
How the recommended device works.
3. The method of claim 2,
Each of the user matrix and the keyword matrix corresponds to a diagonal matrix,
How the recommended device works.
delete delete According to claim 1,
The determining step is
Determining the keyword corresponding to the largest predicted probability value among the predicted probability values for the individual user as the preferred keyword of the individual user
containing,
How the recommended device works.
According to claim 1,
recommending content related to the determined preferred keyword to one or more of the users;
further comprising,
How the recommended device works.
Iteratively learns a first feature vector for users and a second feature vector for keywords to generate an optimal keyword selection prediction matrix for the users and the keywords, and based on the generated optimal keyword selection prediction matrix determines a preferred keyword for one or more of the users, and provides the determined preferred keyword to one or more of the users
including,
The processor is
generating a first matrix indicating whether each of the users has selected each of the keywords, generating a second matrix indicating how much each of the users prefers each of the keywords, the first feature vector and the calculating a first keyword selection prediction matrix using a second feature vector, calculating a difference between the first matrix and the calculated first keyword selection prediction matrix, and multiplying the second matrix and the calculated difference by a matrix to determine an error matrix, and learn the first feature vector and the second feature vector using the determined error matrix,
The generated optimal keyword selection prediction matrix includes a component (n, m), wherein the component (n, m) represents a predicted probability value at which the nth user among the users will select the mth keyword among the keywords,
Recommended device.
10. The method of claim 9,
The processor is
The information on the users is converted into a user matrix and the information on the keywords is converted into a keyword matrix, and the first feature vector and the second feature vector are converted based on the converted user matrix and the converted keyword matrix. to create,
Recommended device.
11. The method of claim 10,
A matrix in which information on content related to each of the keywords is converted is added to the keyword matrix,
Recommended device.
11. The method of claim 10,
Each of the user matrix and the keyword matrix corresponds to a diagonal matrix,
Recommended device.
delete delete 10. The method of claim 9,
The processor is
determining, as the preferred keyword of the individual user, a keyword corresponding to the largest predicted probability value among the predicted probability values for the individual user;
Recommended device.
10. The method of claim 9,
The processor is
recommending content related to the determined preferred keyword to one or more of the users;
Recommended device.

Images