KR20210142484A - Method and Apparatus for Recommendation of Similar Content Items - Google Patents

Abstract

A method for recommending content items is provided. The method includes the steps of: selecting a specific content among at least one content viewed by a user; selecting a target content; calculating a content-based collaborative filtering (CF) similarity between the specific content and the target content; calculating a keyword similarity between the specific content and the target content; calculating a final similarity by weighted average of the CF similarity and the keyword similarity; and determining whether to recommend the target content to the user based on the final similarity.

Description

Method and Apparatus for Recommendation of Similar Content Items}

The present invention relates to information processing technology related to video on demand (VOD) content.

In recent years, as the use of the Internet has become commonplace, the era of home networks has arrived. A specific example of realizing such a home network is an Internet Protocol Television (IPTV) service. The IPTV service is an interactive TV service provided using the Internet, and is connected to a content providing server operated by a content providing service provider using a set-top box connected to the Internet, such as video on demand (VOD) content. It is a service that allows you to download and watch content in a download or streaming manner. Unlike general cable broadcasting, IPTV service provides a variety of attractions and convenience to viewers in that it allows viewers to select and watch the programs they want to watch at a convenient time. It is becoming a revenue model that can increase sales. Therefore, from the perspective of IPTV service providers, marketing various contents to viewers in various ways to encourage purchases is of great interest.

SUMMARY OF THE INVENTION An object of the present invention is to provide a similar content recommendation method and apparatus in which a user can determine and recommend previously viewed content and similar content corresponding to a common genre and/or material.

The problems to be solved by the present invention are not limited to the problems mentioned above, and other problems not mentioned will be clearly understood by those skilled in the art from the following description.

In one aspect, a content recommendation method is provided. The method includes the steps of selecting a specific content among at least one content viewed by a user, selecting a target content, and calculating a content-based collaborative filtering (CF) similarity between the specific content and the target content. , calculating a keyword similarity between the specific content and the target content, calculating a final similarity by weighted average of the CF similarity and the keyword similarity, and recommending the target content to the user based on the final similarity It may include the step of determining whether or not

In an embodiment, calculating the content-based collaborative filtering (CF) similarity between the specific content and the target content includes the following formula

는 w와 Wi의 합집합(union)을 나타내고,

는 w와 Wi의 합집합의 원소들의 개수를 나타내고,

는 w와 Wi의 교집합(intersection)을 나타내고,

는 w와 Wi의 교집합의 원소들의 개수를 나타내고, x_i는 상기 특정 컨텐츠와 컨텐츠 i간의 CF 유사도를 나타냄 - 에 따라 상기 CF 유사도를 산출하는 단계를 포함한다.- Here, w represents a set of IDs (Identifications) of users who have viewed the specific content, and W _i represents a set of IDs of users who have viewed content i,

represents the union of w and Wi,

represents the number of elements of the union of w and Wi,

represents the intersection of w and Wi,

and represents the number of elements of the intersection of w and Wi, and x _i represents the CF similarity between the specific content and the content i.

In an embodiment, calculating the keyword similarity between the specific content and the target content includes referring to a keyword vector for the specific content and a keyword vector for the target content.

In one embodiment, the keyword vector for the specific content has a number of elements corresponding to the total number of candidate keywords, the elements respectively correspond to the candidate keywords, and each of the elements is the number of elements in the specific content. It has an element value of a size corresponding to the importance of the corresponding candidate keyword, and each of the elements takes a value greater than or equal to 0 and less than or equal to the maximum value.

In an embodiment, when an element value of a specific element of the keyword vector for the specific content is 0, it indicates that the candidate keyword corresponding to the specific element is not related to the specific content.

In one embodiment, the keyword vector for the target content has a number of elements corresponding to the total number of candidate keywords, the elements respectively corresponding to the candidate keywords, and each of the elements in the target content It has an element value of a size corresponding to the importance of the corresponding candidate keyword, and each of the elements takes a value greater than or equal to 0 and less than or equal to the maximum value.

In an embodiment, when an element value of a specific element of the keyword vector for the target content is 0, it indicates that the candidate keyword corresponding to the specific element is not related to the target content.

In an embodiment, calculating the keyword similarity between the specific content and the target content includes the following formula

는 상기 특정 컨텐츠에 대한 키워드 벡터 k와 컨텐츠 i에 대한 키워드 벡터 K_i 간의 맨하탄 거리(Manhattan distance)이고, MAX는 상기 최고값에 상기 후보 키워드들의 총 개수를 곱한 값이고, y_i는 상기 특정 컨텐츠와 컨텐츠 i간의 키워드 유사도를 나타냄 - 에 따라 상기 특정 컨텐츠와 상기 타겟 컨텐츠 간의 키워드 유사도를 산출하는 단계를 포함한다.- where k is a keyword vector for the specific content, K _i is a keyword vector for content i,

is the Manhattan distance between the keyword vector k for the specific content and the keyword vector K _i for the content i, MAX is the maximum value multiplied by the total number of the candidate keywords, and y _i is the specific content and calculating the keyword similarity between the specific content and the target content according to - indicating the keyword similarity between the content i and the content i.

In an embodiment, calculating the final similarity by weighted average of the CF similarity and the keyword similarity includes the following equation

와

는 0 이상이고 1이하인 값이고,

와

의 합은 1이고,

는 상기 특정 컨텐츠에 대한 키워드 벡터의 맨하탄 거리이고,

는 컨텐츠 i에 대한 키워드 벡터의 맨하탄 거리이고,

는 키워드 필터 함수이고, S_i는 상기 특정 컨텐츠와 컨텐츠 i간의 최종 유사도를 나타냄 - 에 따라 상기 최종 유사도를 산출하는 단계를 포함한다.- here

Wow

is a value greater than or equal to 0 and less than or equal to 1,

Wow

the sum of is 1,

is the Manhattan distance of the keyword vector for the specific content,

is the Manhattan distance of the keyword vector for content i,

is a keyword filter function, and S _i represents the final similarity between the specific content and the content i - calculating the final similarity.

는 0의 값을 갖는다.In one embodiment, when elements having a non-zero value in the keyword vector for the specific content and elements having a non-zero value in the keyword vector for the target content are mutually exclusive, the specific content and the target content are indicates that none of the candidate keywords are shared and the keyword filter function

has a value of 0.

In an embodiment, the determining whether to recommend the target content to the user based on the final similarity includes determining that the target content is recommended to the user when the final similarity is greater than or equal to a predetermined threshold. includes steps.

In another aspect, a content recommendation method is provided. The method includes the steps of selecting a specific content from among at least one content viewed by a user, selecting a plurality of target content, calculating final similarities between the specific content and the plurality of target content, and the final similarity The method may include determining to recommend at least one target content among the plurality of target contents to the user based on the information. Each of the final similarities may be calculated based on a content-based collaborative filtering (CF) similarity between the specific content and the corresponding target content and a keyword similarity between the specific content and the corresponding target content.

In an embodiment, each of the final similarities is calculated by averaging the CF similarity and the keyword similarity.

In another aspect, an apparatus for content recommendation is provided. The apparatus includes a content table and a database unit for storing keyword vectors for the contents, in which the contents table stores IDs of users who have viewed the corresponding contents in association with each of the contents, and a processing engine. can do. The processing engine refers to an operation of calculating a content-based collaborative filtering (CF) similarity between the specific content viewed by the user and the target content with reference to the content table, a keyword vector for the specific content and the target content. Calculating a keyword similarity between the specific content and the target content using a keyword vector, calculating a final similarity by weighted average of the CF similarity and the keyword similarity, and selecting the target content based on the final similarity to the user It may be configured to perform an operation of determining whether to recommend to the user.

In one embodiment, the processing engine is

는 w와 Wi의 합집합(union)을 나타내고,

는 w와 Wi의 합집합의 원소들의 개수를 나타내고,

는 w와 Wi의 교집합(intersection)을 나타내고,

는 w와 Wi의 교집합의 원소들의 개수를 나타내고, x_i는 상기 특정 컨텐츠와 컨텐츠 i간의 CF 유사도를 나타냄 - 에 따라 상기 CF 유사도를 산출하는 동작을 수행하도록 더 구성된다.- Here, w represents a set of IDs of users who have viewed the specific content, and W _i represents a set of IDs of users who have viewed content i,

represents the union of w and Wi,

represents the number of elements of the union of w and Wi,

represents the intersection of w and Wi,

represents the number of elements of the intersection of w and Wi, and x _i represents the CF similarity between the specific content and the content i.

In one embodiment, the keyword vector for the specific content has a number of elements corresponding to the total number of candidate keywords, the elements respectively correspond to the candidate keywords, and each of the elements is the number of elements in the specific content. It has an element value of a size corresponding to the importance of the corresponding candidate keyword, and each of the elements takes a value greater than or equal to 0 and less than or equal to the maximum value.

In an embodiment, when an element value of a specific element of the keyword vector for the specific content is 0, it indicates that the candidate keyword corresponding to the specific element is not related to the specific content.

In one embodiment, the keyword vector for the target content has a number of elements corresponding to the total number of candidate keywords, the elements respectively corresponding to the candidate keywords, and each of the elements in the target content It has an element value of a size corresponding to the importance of the corresponding candidate keyword, and each of the elements takes a value greater than or equal to 0 and less than or equal to the maximum value.

In an embodiment, when an element value of a specific element of the keyword vector for the target content is 0, it indicates that the candidate keyword corresponding to the specific element is not related to the target content.

In one embodiment, the processing engine is

는 상기 특정 컨텐츠에 대한 키워드 벡터 k와 컨텐츠 i에 대한 키워드 벡터 K_i 간의 맨하탄 거리(Manhattan distance)이고, MAX는 상기 최고값에 상기 후보 키워드들의 총 개수를 곱한 값이고, y_i는 상기 특정 컨텐츠와 컨텐츠 i간의 키워드 유사도를 나타냄 - 에 따라 상기 특정 컨텐츠와 상기 타겟 컨텐츠 간의 키워드 유사도를 산출하는 동작을 수행하도록 더 구성된다.- where k is a keyword vector for the specific content, K _i is a keyword vector for content i,

is the Manhattan distance between the keyword vector k for the specific content and the keyword vector K _i for the content i, MAX is the maximum value multiplied by the total number of the candidate keywords, and y _i is the specific content and indicates the keyword similarity between the content i and - according to the operation of calculating the keyword similarity between the specific content and the target content.

In one embodiment, the processing engine is

와

는 0 이상이고 1이하인 값이고,

와

의 합은 1이고,

는 상기 특정 컨텐츠에 대한 키워드 벡터의 맨하탄 거리이고,

는 컨텐츠 i에 대한 키워드 벡터의 맨하탄 거리이고,

는 키워드 필터 함수이고, S_i는 상기 특정 컨텐츠와 컨텐츠 i간의 최종 유사도를 나타냄 - 에 따라 상기 최종 유사도를 산출하는 동작을 수행하도록 더 구성된다.- here

Wow

is a value greater than or equal to 0 and less than or equal to 1,

Wow

the sum of is 1,

is the Manhattan distance of the keyword vector for the specific content,

is the Manhattan distance of the keyword vector for content i,

is a keyword filter function, and S _i represents a final similarity between the specific content and the content i.

는 0의 값을 갖는다.In one embodiment, when elements having a non-zero value in the keyword vector for the specific content and elements having a non-zero value in the keyword vector for the target content are mutually exclusive, the specific content and the target content are indicates that none of the candidate keywords are shared and the keyword filter function

has a value of 0.

In an embodiment, the processing engine is further configured to perform an operation of determining to recommend the target content to the user when the final similarity is equal to or greater than a predetermined threshold.

In another aspect, an apparatus for content recommendation is provided. The apparatus may include a database unit for storing content-related data, and a processing engine. The processing engine calculates final similarities between the specific content viewed by the user and a plurality of target contents with reference to the content-related data, and selects at least one target content from among the plurality of target contents based on the final similarities and determine a recommendation to a user, wherein the processing engine is configured to determine each of the final similarities between the specific content and the corresponding target content and a content-based collaborative filtering (CF) similarity between the specific content and the corresponding target content and the specific content and the corresponding target content. It may be further configured to calculate based on the degree of similarity between keywords.

In one embodiment, the content-related data includes data about keyword vectors for the content table and the content, and the content table is associated with IDs of users who viewed the content for each of the content and stored. do.

In an embodiment, the processing engine is further configured to calculate each of the final similarities by weighted average of the CF similarity and the keyword similarity.

In another aspect, a computer-readable recording medium recording a program is provided. wherein the program includes instructions, the instructions performing the method when executed by a computer.

According to embodiments of the present invention, content is determined and recommended by a user based on the similarity of previously viewed content and similar content corresponding to a common genre and/or material by content-based collaborative filtering and keyword similarity. There is a technological effect that can increase consumption.

1 is a diagram illustrating an embodiment of a block diagram of an apparatus for recommending similar contents.
2 is a diagram illustrating an embodiment of a content table.
3 is a diagram illustrating an embodiment of a flowchart for explaining a method for recommending similar content.

Advantages and features of the present invention and a method of achieving them will become apparent with reference to the embodiments described below in detail in conjunction with the accompanying drawings. However, the present invention is not limited to the embodiments disclosed below, but will be implemented in a variety of different forms, and these embodiments merely allow the disclosure of the present invention to be complete and those of ordinary skill in the art to which the present invention pertains. It is provided to fully inform the person of the scope of the invention, and the present invention is only defined by the scope of the claims.

The terms used herein are used only to describe specific embodiments and are not intended to limit the present invention. For example, a component expressed in a singular should be understood as a concept including a plurality of components unless the context clearly means only the singular. In addition, in the specification of the present invention, terms such as 'comprise' or 'have' are only intended to designate that the features, numbers, steps, operations, components, parts, or combinations thereof described in the specification exist, and such The use of the term does not exclude the possibility of the presence or addition of one or more other features or numbers, steps, operations, components, parts, or combinations thereof. Also, in the embodiments described in this specification, a 'module' or a 'unit' may mean a functional part that performs at least one function or operation.

In addition, 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 this invention belongs. Terms such as those defined in a commonly used dictionary should be interpreted as having a meaning consistent with the meaning in the context of the related art, and unless explicitly defined in the specification of the present invention, it should be interpreted in an ideal or excessively formal meaning. doesn't happen

Hereinafter, embodiments of the present invention will be described in more detail with reference to the accompanying drawings. However, in the following description, if there is a risk of unnecessarily obscuring the gist of the present invention, detailed descriptions of well-known functions or configurations will be omitted.

1 is a diagram illustrating an embodiment of a block diagram of an apparatus for recommending similar contents.

The similar content recommendation apparatus 100 of FIG. 1 may be an example of a system implemented as computer programs operated by an IPTV service provider and executed on one or more server computers installed in one or more locations. As shown in FIG. 1 , the similar content recommendation apparatus 100 may include a database unit 110 and a processing engine 120 communicatively coupled to the database unit 110 . The database unit 110 may store content-related data. The content-related data may include data about the content table. In the content table, IDs of users who have viewed the corresponding content for each of the content are stored in association with each other. In this specification, the meaning that the user watched the content should be interpreted as encompassing the case where the user watched the trailer of the content or purchased the content and watched the main part. Referring to FIG. 2 showing an embodiment of a content table, IDs (Identifications) of users who have viewed or inquired corresponding content for each content are listed. According to the illustrated embodiment, content 1 is recorded as viewed or viewed by user 10 , user 33 , user 72 , and user 165 . In addition, content 3 is recorded as viewed or viewed by user 10, user 72, user 165, and user 289. The content table 200 shown in FIG. 2 is only conceptually illustrating a method of storing data in association with each other for the purpose of illustration, and the contents shown in FIG. 2 do not illustrate or imply the structure of the stored data. will have to recognize

The data about the content may further include data about keyword vectors for the content. A keyword vector for any content has a number of elements corresponding to the total number of candidate keywords. Here, the candidate keywords may be a total set of keywords related to contents provided by the IPTV service. In one embodiment, the candidate keyword is a meta keyword associated with the content. The elements of the keyword vector respectively correspond to or are associated with the candidate keywords. Each of the elements of the keyword vector may have an element value having a size corresponding to the importance of the corresponding candidate keyword in the corresponding content. Each of the elements of the keyword vector may take a value greater than or equal to zero and less than or equal to the highest value. For example, assuming that the total number of candidate keywords is 5, the highest value is 10, and the keyword vector for content 1 is v ₁ = [7, 1, 9, 0, 2], The importance is 7, the importance of the candidate keyword No. 2 is 1, the importance of the candidate keyword No. 3 is 9, the importance of the candidate keyword No. 5 is 2, and the importance of the candidate keyword No. 4 is not related to the specific content above. As another example, assuming that the keyword vector for content 2 is v ₂ = [1, 3, 0, 0, 10], the importance of candidate keyword #1 in content #2 is 1, and the importance of candidate keyword #2 is 3, the importance of candidate keyword No. 5 is 10, which is the highest level, and candidate keywords No. 3 and No. 4 are not related to content No. 2. As another example, assuming that the keyword vector for content #3 is v ₃ = [0, 0, 0, 4, 0], the importance of candidate keyword #4 in content #3 is 4, and the remaining candidate keywords are That means it has nothing to do with number 3.

The database unit 110 may further store software/firmware necessary to implement the processing engine 120 . The database unit 110, a flash memory type (flash memory type), a hard disk type (hard disk type), a multimedia card (MultiMedia Card: MMC), a card type memory (eg, SD (Secure Digital) card or XD (eXtream Digital) cards, etc.), Random Access Memory (RAM), Static Random Access Memory (SRAM), Read-Only Memory (ROM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Programmable Read-Only Memory (PROM) ), a magnetic memory, a magnetic disk, and an optical disk may be implemented as any one storage medium, but those skilled in the art will understand that the implementation form of the database unit 110 is not limited thereto.

The processing engine 120 may be designed to select a specific content from among at least one content viewed by the user and determine that other content having a high similarity to the specific content is recommended to the user. To this end, the processing engine 120 may be configured to select any one target content excluding the specific content viewed by the user from among the contents provided by the IPTV service and compare it with the specific content viewed by the user. The processing engine 120 refers to the content table 200 to calculate a content-based collaborative filtering (CF) similarity between a specific content and a target content, and a keyword vector for a specific content and a target content. The operation of calculating the keyword similarity between the specific content and the target content using a keyword vector, the operation of calculating the final similarity by weighted average of the CF similarity and the keyword similarity, and determining whether to recommend the target content to the user based on the final similarity It may be configured to perform an operation.

The processing engine 120 may be further configured to perform an operation of calculating a CF similarity between specific content and target content according to Equation 1 below.

는 w와 Wi의 합집합(union)을 나타내고,

는 w와 Wi의 합집합의 원소들의 개수를 나타내고,

는 w와 Wi의 교집합(intersection)을 나타내고,

는 w와 Wi의 교집합의 원소들의 개수를 나타내고, x_i는 특정 컨텐츠와 컨텐츠 i간의 CF 유사도를 나타낸다.Here, w denotes a set of IDs of users who watched a specific content, W _i denotes a set of IDs of users who watched content i,

represents the union of w and Wi,

represents the number of elements of the union of w and Wi,

represents the intersection of w and Wi,

denotes the number of elements of the intersection of w and Wi, and x _i denotes the CF similarity between specific content and content i.

For example, referring to the content table 200 of FIG. 2 , the set of users who watched content 1 is C ₁ ={10, 33, 72, 165}, and the set of users who watched content number 4 is Since C ₄ = {72, 162, 555}, the CF similarity between content number 1 (specific content) and content number 4 (target content) x ₄ = 1 / 6 = 0.16. As another example, the set of users who watched content 1 is C ₁ ={10, 33, 72, 165}, and the set of users who watched content 2 is C ₂ = {10, 92, 165} Therefore, the CF similarity between Content No. 1 (specific content) and Content No. 2 (target content) x ₂ = 2 / 5 = 0.4. As another example, the set of users who watched content 1 is C ₁ ={10, 33, 72, 165}, and the set of users who watched content 3 is C ₃ = {10, 72, 165 , 289} so, content # 1 as the degree of similarity CF x ₃ = _3/5 = 0.6 between (the specific content) and the content # 3 (the target content) has a relatively high value. This is because most of the users who watched the content 1 and the content 3 are similar.

The processing engine 120 may be further configured to perform an operation of calculating a keyword similarity between specific content and target content according to Equation 2 below.

는 특정 컨텐츠에 대한 키워드 벡터 k와 컨텐츠 i에 대한 키워드 벡터 K_i 간의 맨하탄 거리(Manhattan distance)이고, MAX는 키워드 벡터의 원소들이 취할 수 있는 최고값에 후보 키워드들의 총 개수를 곱한 값이고, y_i는 특정 컨텐츠와 컨텐츠 i간의 키워드 유사도를 나타낸다.where k is a keyword vector for specific content, K _i is a keyword vector for content i,

is the Manhattan distance between the keyword vector k for a specific content and the keyword vector K _i for the content i, MAX is the maximum value that elements of the keyword vector can take multiplied by the total number of candidate keywords, y _i represents the keyword similarity between specific content and content i.

As in the above example, the maximum value that the elements of the keyword vector can take is 10, the total number of candidate keywords is 5, the keyword vector for content 1 is v ₁ = [7, 1, 9, 0, 2], A keyword in the case where the keyword vector for content 2 is v ₂ = [1, 3, 0, 0, 10] and the keyword vector for content 3 is v ₃ = [0, 0, 0, 4, 0] _{As an example of calculating the similarity, the keyword similarity y 2} = 1 - 25/50 = 0.5 between the first content (specific content) and the second content (target content). _{As another example, the keyword similarity y 3} = 1 - 23/50 = 0.54 between content number 1 (specific content) and content number 3 (target content).

The processing engine 120 may be further configured to perform an operation of calculating a final similarity according to Equations 3 and 4 below.

와

는 0 이상이고 1이하인 값이고,

와

의 합은 1이다. 일 실시예에서,

와

를 모두 0.5로 설정하여 CF 유사도와 키워드 유사도에 동일한 가중치를 곱하여 최종 유사도를 산출한다. 일 실시예에서,

보다

를 더 큰 값으로 설정하여 CF 유사도 보다 키워드 유사도에 더 큰 가중치를 곱하여 최종 유사도를 산출한다. 위 수학식 4에서

는 특정 컨텐츠에 대한 키워드 벡터의 맨하탄 거리이고,

는 컨텐츠 i에 대한 키워드 벡터의 맨하탄 거리이다. 위 수학식 3에서 S_i는 특정 컨텐츠와 컨텐츠 i간의 최종 유사도를 나타낸다. 위 수학식 4에서

는 키워드 필터 함수로서 특정 컨텐츠와 컨텐츠 i가 어떠한 후보 키워드도 공유하지 않는 경우 S_i를 0으로 만들기 위한 함수이다.In Equation 3 above

Wow

is a value greater than or equal to 0 and less than or equal to 1,

Wow

The sum of is 1. In one embodiment,

Wow

are set to 0.5, and the final similarity is calculated by multiplying the CF similarity and the keyword similarity by the same weight. In one embodiment,

see

is set to a larger value, and the final similarity is calculated by multiplying the keyword similarity by a larger weight than the CF similarity. In Equation 4 above

is the Manhattan distance of the keyword vector for a specific content,

is the Manhattan distance of the keyword vector for content i. In Equation 3 above, S _i represents the final similarity between specific content and content i. In Equation 4 above

is a keyword filter function, and is a function for making _{S i} 0 when a specific content and content i do not share any candidate keywords.

는 1이고, x₂ = 0.4이고 y₂ = 0.5이므로,

와

가 모두 0.5인 경우 컨텐츠 1번(특정 컨텐츠)과 컨텐츠 2번(타겟 컨텐츠) 간의 최종 유사도 S₂ = 0.45이다. 다른 예를 들어 보면, 컨텐츠 1번에 대한 키워드 벡터 v₁ = [7, 1, 9, 0, 2]이고 컨텐츠 3번에 대한 키워드 벡터 v₃ = [0, 0, 0, 4, 0]이므로, 컨텐츠 1번과 컨텐츠 3번은 어떠한 후보 키워드도 공유하지 않고 이에 따라 키워드 필터 함수

가 0이므로, 컨텐츠 1번(특정 컨텐츠)과 컨텐츠 3번(타겟 컨텐츠) 간의 최종 유사도 S₃은 x₃과 y₃의 값에 상관 없이 0의 값을 갖게 된다. 이와 같이 특정 컨텐츠에 대한 키워드 벡터에서 0이 아닌 값을 갖는 원소들과 타겟 컨텐츠에 대한 키워드 벡터에서 0이 아닌 값을 갖는 원소들이 서로 배타적인 경우 특정 컨텐츠와 타겟 컨텐츠는 후보 키워드들 중 어떠한 후보 키워드도 공유하지 않는 것을 나타내며 이 경우 키워드 필터 함수

를 0으로 만들어 특정 컨텐츠와 타겟 컨텐츠 간의 최종 유사도를 0으로 결정한다.As an example of calculating the final similarity using the case of the above example, the keyword vector v ₁ for content 1 = [7, 1, 9, 0, 2] and the keyword vector v ₂ for content 2 = [1] , 3, 0, 0, 10], so the keyword filter function

is 1, since x ₂ = 0.4 and y ₂ = 0.5,

Wow

_{If both are 0.5, the final similarity S 2} = 0.45 between content 1 (specific content) and content 2 (target content). As another example, since the keyword vector v ₁ for content 1 = [7, 1, 9, 0, 2] and the keyword vector v ₃ for content 3 = [0, 0, 0, 4, 0], , content #1 and content #3 do not share any candidate keywords, and thus the keyword filter function

_{Since is 0, the final similarity S 3} between content 1 (specific content) and content 3 (target content) has a value of 0 regardless of the values of _{x 3} and y _{3 .} As described above, when elements having a non-zero value in the keyword vector for the specific content and elements having a non-zero value in the keyword vector for the target content are mutually exclusive, the specific content and the target content are selected as any candidate keyword among the candidate keywords. also indicates not to share, in this case the keyword filter function

is 0 to determine the final similarity between the specific content and the target content as 0.

The processing engine 120 may be further configured to perform an operation of determining that the target content is recommended to the user when the final similarity is greater than or equal to a predetermined threshold. By setting the threshold low or high, the processing engine 120 can be designed to make decisions about aggressively or conservatively recommending content to a user.

In the above description, the processing engine 120 is configured to determine whether to recommend the target content to the user by selecting one target content and comparing it with the specific content. It is possible to configure the processing engine 120 to calculate and determine to recommend at least one target content among a plurality of target content to the user based on the final similarities.

The processing engine 120 includes Application Specific Integrated Circuits (ASICs), Digital Signal Processors (DSPs), Digital Signal Processing Devices (DSPDs), and programmable logic devices (Programmable Logic Devices). Hardware based on at least one of Logic Devices (PLDs), Field-Programmable Gate Arrays (FPGAs), processors, controllers, micro-controllers, and microprocessors. It may be implemented as a hardware platform. The processing engine 120 may also be implemented as a firmware/software module executable on the hardware platform described above. In this case, the software module may be implemented by a software application written in an appropriate program language.

3 is a diagram illustrating an embodiment of a flowchart for explaining a method for recommending similar content.

As shown in FIG. 3 , the present similar content recommendation method starts with the step of selecting a specific content from among at least one content that the user has watched ( S305 ). In step S310, target content is selected. The target content may be selected as any one content other than the above specific content among content provided by the IPTV service. In step S315, a CF similarity between specific content and target content is calculated. The CF similarity may be calculated according to Equation 1 above. In step S320, the keyword similarity between specific content and target content is calculated. The keyword similarity may be calculated according to Equation 2 above with reference to the keyword vector for the specific content and the keyword vector for the target content. In step S325, the final similarity is calculated by averaging the CF similarity and the keyword similarity. The final similarity may be calculated according to Equations 3 and 4 above. In step S330, it is determined whether to recommend the target content to the user based on the final similarity. In this step, when the final similarity is greater than or equal to a predetermined threshold, it may be determined that the target content is recommended to the user.

In the above description, the meaning of the description that a component is connected to or coupled to another component means that the component is directly connected or coupled to the other component, as well as one or more other components interposed therebetween. It should be understood to include the meaning that may be connected or coupled through an element. In addition, terms for describing the relationship between elements (eg, 'between', 'between', etc.) should also be interpreted with similar meanings.

In the embodiments disclosed herein, the arrangement of the illustrated components may vary depending on the environment or requirements in which the invention is implemented. For example, some components may be omitted or some components may be integrated and implemented as one. Also, the arrangement order and connection of some components may be changed.

In the above, various embodiments of the present invention have been shown and described, but the present invention is not limited to the specific embodiments described above, and the above-described embodiments depart from the gist of the present invention as claimed in the appended claims. Without this, various modifications may be made by those of ordinary skill in the art to which the present invention pertains, and these modified embodiments should not be understood separately from the technical spirit or scope of the present invention. Accordingly, the technical scope of the present invention should be defined only by the appended claims.

100: content recommendation device
110: database unit
120: processing engine
200: content table

Claims (28)

As a content recommendation method,
selecting a specific content from among at least one content watched by the user;
selecting target content;
Calculating a content-based collaborative filtering (CF) similarity between the specific content and the target content;
calculating a keyword similarity between the specific content and the target content;
calculating a final similarity by weighted average of the CF similarity and the keyword similarity; and
and determining whether to recommend the target content to the user based on the final similarity.
According to claim 1,
Calculating the content-based collaborative filtering (CF) similarity between the specific content and the target content includes the following formula

- Here, w represents a set of IDs (Identifications) of users who have viewed the specific content, and W _i represents a set of IDs of users who have viewed content i,

represents the union of w and Wi,

represents the number of elements of the union of w and Wi,

represents the intersection of w and Wi,

represents the number of elements of the intersection of w and Wi, and x _i represents the CF similarity between the specific content and the content i - calculating the CF similarity according to .
3. The method of claim 2,
Calculating the keyword similarity between the specific content and the target content includes referring to a keyword vector for the specific content and a keyword vector for the target content.
4. The method of claim 3,
The keyword vector for the specific content has a number of elements corresponding to the total number of candidate keywords, the elements respectively correspond to the candidate keywords, and each of the elements is the number of elements of the corresponding candidate keyword in the specific content. A method for recommending content, wherein each of the elements has a value of a size corresponding to importance, and each of the elements takes a value greater than or equal to 0 and less than or equal to a maximum value.
5. The method of claim 4,
When an element value of a specific element of the keyword vector for the specific content is 0, it indicates that the candidate keyword corresponding to the specific element is not related to the specific content.
4. The method of claim 3,
The keyword vector for the target content has a number of elements corresponding to the total number of candidate keywords, the elements respectively correspond to the candidate keywords, and each of the elements is the number of elements of the corresponding candidate keyword in the target content. A method for recommending content, wherein each of the elements has a value of a size corresponding to importance, and each of the elements takes a value greater than or equal to 0 and less than or equal to a maximum value.
7. The method of claim 6,
When an element value of a specific element of the keyword vector for the target content is 0, it indicates that the candidate keyword corresponding to the specific element is not related to the target content.
7. The method of claim 4 or 6,
Calculating the keyword similarity between the specific content and the target content includes the following formula

- where k is a keyword vector for the specific content, K _i is a keyword vector for content i,

is the Manhattan distance between the keyword vector k for the specific content and the keyword vector K _i for the content i, MAX is the maximum value multiplied by the total number of the candidate keywords, and y _i is the specific content and calculating the keyword similarity between the specific content and the target content according to - indicating the keyword similarity between the content i and the content i.
9. The method of claim 8,
The step of calculating the final similarity by weighted average of the CF similarity and the keyword similarity is as follows:

- here

Wow

is a value greater than or equal to 0 and less than or equal to 1,

Wow

the sum of is 1,

is the Manhattan distance of the keyword vector for the specific content,

is the Manhattan distance of the keyword vector for content i,

is a keyword filter function, and S _i represents the final similarity between the specific content and the content i - calculating the final similarity according to .
10. The method of claim 9,
When elements having a non-zero value in the keyword vector for the specific content and elements having a non-zero value in the keyword vector for the target content are mutually exclusive, the specific content and the target content are selected from among the candidate keywords. Indicates that no candidate keywords are shared and the keyword filter function

has a value of 0, the content recommendation method.
According to claim 1,
The determining whether to recommend the target content to the user based on the final similarity includes determining that the target content is recommended to the user when the final similarity is greater than or equal to a predetermined threshold value, How to recommend content.
As a content recommendation method,
selecting a specific content from among at least one content watched by the user;
selecting a plurality of target contents;
calculating final similarities between the specific content and the plurality of target content; and
determining to recommend at least one target content among the plurality of target contents to the user based on the final similarities;
Each of the final similarities is calculated based on a content-based collaborative filtering (CF) similarity between the specific content and the corresponding target content and a keyword similarity between the specific content and the corresponding target content.
13. The method of claim 12,
Each of the final similarities is calculated by weighted average of the CF similarity and the keyword similarity.
A device for content recommendation, comprising:
A database unit for storing keyword vectors for a content table and contents, wherein IDs of users who have viewed the corresponding contents are stored in association with each of the contents in the contents table; and
a processing engine;
The processing engine refers to an operation of calculating a content-based collaborative filtering (CF) similarity between the specific content viewed by the user and the target content with reference to the content table, a keyword vector for the specific content and the target content. Calculating a keyword similarity between the specific content and the target content using a keyword vector, calculating a final similarity by weighted average of the CF similarity and the keyword similarity, and selecting the target content based on the final similarity to the user A content recommendation device, configured to perform an operation of determining whether to recommend to a user.
15. The method of claim 14,
The processing engine is

- Here, w represents a set of IDs of users who have viewed the specific content, and W _i represents a set of IDs of users who have viewed content i,

represents the union of w and Wi,

represents the number of elements of the union of w and Wi,

represents the intersection of w and Wi,

represents the number of elements of the intersection of w and Wi, and x _i represents the CF similarity between the specific content and the content i.
16. The method of claim 15,
The keyword vector for the specific content has a number of elements corresponding to the total number of candidate keywords, the elements respectively correspond to the candidate keywords, and each of the elements is the number of elements of the corresponding candidate keyword in the specific content. A content recommendation apparatus having an element value of a size corresponding to importance, wherein each of the elements takes a value greater than or equal to 0 and less than or equal to a maximum value.
17. The method of claim 16,
When an element value of a specific element of the keyword vector for the specific content is 0, it indicates that the candidate keyword corresponding to the specific element is not related to the specific content.
16. The method of claim 15,
The keyword vector for the target content has a number of elements corresponding to the total number of candidate keywords, the elements respectively correspond to the candidate keywords, and each of the elements is the number of elements of the corresponding candidate keyword in the target content. A content recommendation apparatus having an element value of a size corresponding to importance, wherein each of the elements takes a value greater than or equal to 0 and less than or equal to a maximum value.
19. The method of claim 18,
When an element value of a specific element of the keyword vector for the target content is 0, it indicates that the candidate keyword corresponding to the specific element is not related to the target content.
19. The method of claim 16 or 18,
The processing engine is

- where k is a keyword vector for the specific content, K _i is a keyword vector for content i,

is the Manhattan distance between the keyword vector k for the specific content and the keyword vector K _i for the content i, MAX is the maximum value multiplied by the total number of the candidate keywords, and y _i is the specific content and indicating the keyword similarity between the content i and the content i, further configured to perform an operation of calculating the keyword similarity between the specific content and the target content according to
21. The method of claim 20,
The processing engine is

- here

Wow

is a value greater than or equal to 0 and less than or equal to 1,

Wow

the sum of is 1,

is the Manhattan distance of the keyword vector for the specific content,

is the Manhattan distance of the keyword vector for content i,

is a keyword filter function, and S _i represents a final degree of similarity between the specific content and content i.
22. The method of claim 21,
When elements having a non-zero value in the keyword vector for the specific content and elements having a non-zero value in the keyword vector for the target content are mutually exclusive, the specific content and the target content are selected from among the candidate keywords. Indicates that no candidate keywords are shared and the keyword filter function

has a value of 0, a content recommendation device.
15. The method of claim 14,
The processing engine is further configured to perform an operation of determining to recommend the target content to the user when the final similarity is greater than or equal to a predetermined threshold.
A device for content recommendation, comprising:
a database unit for storing content-related data; and
a processing engine;
The processing engine calculates final similarities between the specific content viewed by the user and a plurality of target contents with reference to the content-related data, and selects at least one target content from among the plurality of target contents based on the final similarities is configured to determine a recommendation to a user;
The processing engine is further configured to calculate each of the final similarities based on a content-based collaborative filtering (CF) similarity between the specific content and the corresponding target content and a keyword similarity between the specific content and the corresponding target content. A content recommendation device.
25. The method of claim 24,
The content-related data includes a content table and data on keyword vectors for the content, and the content table stores IDs of users who have viewed the corresponding content in association with each of the content items. .
25. The method of claim 24,
The processing engine is further configured to calculate each of the final similarities by weighted average of the CF similarity and the keyword similarity.
A computer-readable recording medium recording a program, the program including instructions, which, when executed by a computer, perform the method according to any one of claims 1 to 11.
A computer-readable recording medium recording a program, the program including instructions, which, when executed by a computer, perform the method according to any one of claims 12 and 13.

Images