KR102150660B1 - Method to recommend digital contents based on search log and apparatus therefor - Google Patents

Abstract

The present specification discloses a content recommendation device. The device comprises: a search history analysis unit for extracting one or more of a search keyword and selected content information selected by the user from a search result according to the search keyword from a user's content search history; A recommendation score calculator configured to calculate a recommendation score for at least one recommendation candidate content based on at least one of the search keyword and the selected content information; And a recommended content determination unit that determines recommended content to be recommended to the user according to the calculated recommendation score.

Description

Method and device for recommending digital contents based on search history {METHOD TO RECOMMEND DIGITAL CONTENTS BASED ON SEARCH LOG AND APPARATUS THEREFOR}

The present specification relates to a content recommendation device and a content recommendation service method, and more particularly, to a method of recommending a content closest to a user's taste based on information obtained from a user's content search history, and a device used therein. .

BACKGROUND ART Recently, as communication networks have developed rapidly, the usage rate of communication services using mobile terminals as well as communication services using existing fixed computers is increasing rapidly, and by this, various types of communication services are becoming very diverse.

Among the communication services for computers and/or mobile terminals, the most representative ones include a content providing service that enables users to purchase various content or application programs such as music, video, and coupons, and an app store.

Recently, as an environment in which general users who are not professional producers can sufficiently produce such contents or applications, the number and types of contents that exist online are increasing innumerably.

Dizzy like this?? As the size of the market that provides content (app, video, audio, eBook) increases, users' choices increase, but it is difficult to find content that is highly relevant to the user's interests or tastes. Therefore, the importance of selecting and recommending content based on the user's usage behavior increases.In the case of the existing recommendation system based on the user's purchase/evaluation history, the recommendation is provided when the user's purchase/evaluation history is lacking or lacking. There is a limit in which the number of content cannot be or is limited. Therefore, as a complementary measure when it is difficult to provide a recommendation service due to insufficient purchase/evaluation information, it is possible to estimate the keyword of interest and the category of interest from the user's search history and selection of the search result, and based on the content, the user's There is a request for a way to recommend content that is highly relevant to your interests.

An object of the present specification is to provide a content recommendation apparatus and method. More specifically, it estimates the keyword of interest and the category of interest from the user's search history and selection of the search results, and recommends the contents that are highly relevant to the user's interests among all contents based on the contents to promote the user's use of the service. It is an object to provide a method and apparatus. In addition, one or more of the selected content information selected by the user is extracted from the search keyword and the search results according to the search keyword from the user's content search history, and at least one recommendation based on at least one of the search keyword and the selected content information It is an object of the present specification to improve user satisfaction by providing a content recommendation device that calculates a recommendation score for candidate content and determines recommended content to be recommended to the user according to the calculated recommendation score.

According to an embodiment of the present specification, a content recommendation device is provided. The device comprises: a search history analysis unit for extracting one or more of a search keyword and selected content information selected by the user from a search result according to the search keyword from a user's content search history; A recommendation score calculator configured to calculate a recommendation score for at least one recommendation candidate content based on at least one of the search keyword and the selected content information; It may include a recommended content determination unit for determining the recommended content to be recommended to the user according to the calculated recommendation score.

The selected content information may include a category of the content selected by the user and a keyword associated with the selected content.

The recommendation score calculation unit may include a first score according to whether the search keyword is included in the detailed information of the recommended candidate content, and a first score according to whether a related keyword of the search keyword is included in the detailed information of the recommended candidate content. 2 score, a third score according to a degree of relevance between the category of the recommended candidate content and the search keyword, a fourth score according to a degree of relevance between the category of the selected content and the category of the recommended candidate content, and the association of the selected content The recommendation score may be calculated based on at least one of a fifth score according to a degree of relevance between a keyword and a related keyword of the recommendation candidate content.

The recommendation score calculator may calculate the recommendation score by giving the second score a weight determined in proportion to the number of related keywords of the search keyword included in the detailed information of the candidate content.

The related keyword of the search keyword is a related keyword whose degree of relevance to the search keyword is greater than or equal to a predetermined criterion, and the weight may be determined by reflecting the number of the related keywords and the degree of relevance of each of the related keywords to the search keyword. .

The recommendation score calculation unit calculates the recommendation score by summing the first score to the fifth score at a predetermined reflection ratio, and the predetermined reflection ratio may be updated according to a recommendation response result of the user.

According to another embodiment of the present specification, a method for recommending content is provided. The method includes the steps of extracting at least one of a search keyword and selected content information selected by the user from a search result according to the search keyword from the user's content search history; Calculating a recommendation score for at least one recommendation candidate content based on at least one of the search keyword and the selected content information; And determining recommended content to be recommended to the user according to the calculated recommendation score.

The selected content information may include a category of the content selected by the user and a keyword associated with the selected content.

The calculating of the recommendation score may include a first score according to whether or not the search keyword is included in the detailed information of the recommendation candidate content, and whether a related keyword of the search keyword is included in the detailed information of the recommendation candidate content. A second score according to, a third score according to a degree of relation between the category of the recommended candidate content and the search keyword, a fourth score according to a degree of relation between the category of the selected content and the category of the recommended candidate content, and the selection It may be a step of calculating the recommendation score based on at least one of a fifth score according to a degree of relation between a related keyword of content and a related keyword of the recommendation candidate content.

The calculating of the recommendation score may be a step of calculating the recommendation score by assigning a weight determined in proportion to the number of related keywords of the search keyword included in the detailed information of the candidate content to the second score.

The related keyword of the search keyword is a related keyword whose degree of relevance to the search keyword is greater than or equal to a predetermined criterion, and the weight may be determined by reflecting the number of the related keywords and the degree of relevance of each of the related keywords to the search keyword. .

The calculating of the recommendation score is a step of calculating the recommendation score by summing the first score to the fifth score at a predetermined reflection ratio, and the predetermined reflection ratio is updated according to the recommendation response result of the user. Can be.

According to another embodiment of the present specification, a content recommendation service system is provided. The system extracts one or more of the selected content information selected by the user from a search keyword and a search result according to the search keyword from the user's content search history, and one or more of the search keyword and the selected content information A content recommendation device that calculates a recommendation score for the above recommendation candidate content and determines recommended content to be recommended to the user according to the calculated recommendation score; It may include a user terminal that displays the recommended content recommended by the content recommendation device.

Embodiments of the present specification have an effect of estimating a keyword of interest and a category of interest from a user's search history and selection of a search result, and recommending content that is highly relevant to the user's interest among all contents based on the corresponding content. Furthermore, it is possible to increase the accuracy of analysis on customer taste by constantly learning the user's preferred category and preferred keywords from the search history, and by providing real-time recommendations according to the time of search and result selection, capturing user's volatile interests and using the service. It can improve the activity.

1 is a block diagram of a content recommendation device according to an embodiment of the present invention.
2 is a block diagram showing a content recommendation service system according to an embodiment of the present invention.
3 is a flowchart illustrating a content recommendation method according to an embodiment of the present invention.

It should be noted that the technical terms used in the present specification are only used to describe specific embodiments, and are not intended to limit the present invention. In addition, the technical terms used in the present specification should be interpreted as generally understood by those of ordinary skill in the technical field to which the present invention belongs, unless otherwise defined in the present specification, and excessively comprehensive It should not be construed as a human meaning or an excessively reduced meaning. In addition, when a technical term used in the present specification is an incorrect technical term that does not accurately express the spirit of the present invention, it will be replaced with a technical term that can be correctly understood by those skilled in the art to be understood. In addition, general terms used in the present invention should be interpreted as defined in the dictionary or according to the context before and after, and should not be interpreted as an excessively reduced meaning.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings, but the same or similar components are assigned the same reference numerals regardless of the reference numerals, and redundant descriptions thereof will be omitted. In addition, in describing the present invention, when it is determined that a detailed description of a related known technology may obscure the subject matter of the present invention, a detailed description thereof will be omitted. In addition, it should be noted that the accompanying drawings are only for easily understanding the spirit of the present invention and should not be construed as limiting the spirit of the present invention by the accompanying drawings. The spirit of the present invention should be construed as extending to all changes, equivalents, or substitutes in addition to the accompanying drawings.

1 is a block diagram of a content recommendation device according to an embodiment of the present invention.

Referring to FIG. 1, the content recommendation device 100 may include a search history analysis unit 101, a recommendation score calculation unit 102, and a recommended content determination unit 103.

The content recommendation device 100 may analyze a search history of digital content of a specific user, analyze a keyword of interest and a category of interest of the corresponding user, and recommend personalized content using the result. In particular, the server may identify the user's interest content, interest keyword, interest category, etc. by analyzing a log (e.g., search keyword, content searched among search results, etc.) . When a user's interest content, interest keyword, and the like are analyzed by the method and apparatus disclosed herein, a recommendation of the content most closely related to the interest may be provided.

In order to efficiently select recommended content according to the characteristics of the user analyzed from the search history, the content recommendation device 100 may perform a preliminary analysis on the available content (recommended candidate content) held. That is, the content recommendation device 100 may perform a task of analyzing/classifying the characteristics of all recommended candidate content so that the user's search history (eg, a specific search keyword) and the mapping (mapping) to determine the recommended content. have.

As an embodiment, the content recommendation device 100 may generate and store association information between a category-keyword, a keyword-keyword, and a keyword-category of recommendation candidate content.

In general, the digital content market includes not only information such as content ID, content name, content price, content provider, etc. for each content, but also detailed information about each content (eg, type, category, airing date, related keywords, etc.). have. Using this, the content recommendation device 100 may generate category information, related keyword information, and the like for each content from the detailed information. At this time, the category information may have a system that allows multiple levels or overlapping according to the nature of the content market. Meanwhile, the related keyword information is primarily extracted through a morpheme analyzer to extract a set of nouns, and keywords that are significant within the set can also be extracted through the TF-IDF (Term Frequency-Inverse Document Frequency) method within the category of the content. have.

In this way, the content recommendation device 100 may generate/classify information such as a content ID, a category ID, and a keyword ID of each content and store the corresponding information. According to the number of keywords extracted from the content, one or more'category-keyword' information may be provided, and a relationship between each keyword and a category, a relationship between each keyword, etc. may be determined and stored. Table 1 below is an example of related information between keywords and categories generated for arbitrary content.

keyword Related Category Category-keyword relevance
K1
C1 0.9 C3 0.1 C4 0.4
K2 C2 0.5 C3 0.6
K3 C1 0.7 C5 0.9

Referring to Table 1, the keywords related to arbitrary content are K1, K2, and K3, and each keyword is related to a plurality of categories, and the degree of relevance of each is analyzed. This analysis information is used to calculate the fitness of recommendations for future content recommendations.

Table 2 below is an example of related information between keywords and keywords generated for arbitrary content.

Keyword-keyword relevance K1 K2 K3 K1 - 0.95 0.12 K2 0.95 - 0.55 K3 0.12 0.55 -

Referring to Table 2, the relevance of each of the keywords K1, K2, and K3 associated with arbitrary content is analyzed. Based on the analysis, if the representative keyword of the content is K1, it may be expected that the keyword K3 has a lower relevance to the content than that of K2. Such analysis information is used to calculate a recommendation fit for future content recommendation.

Thereafter, the search history analysis unit 101 of the content recommendation device 100 may extract one or more of the selected content information selected by the user from a search keyword and a search result according to the search keyword from the content search history of a specific user.

That is, by analyzing the history of a user's search for content on a specific internet site (e.g., content market), (1) information on a search keyword (or search word), and (2) a user from among the search results according to the search keyword. Extracts information on the content selected and retrieved. Hereinafter, the above (2) is referred to as "selected content information". In this case, the selected content information may include an ID of the corresponding content, a category of the corresponding content, detailed information of the corresponding content, a related keyword of the corresponding content, and a degree of a relationship between the related keyword and the category.

Next, the recommendation score calculation unit 102 of the content recommendation device 100 calculates a recommendation score for at least one recommendation candidate content based on at least one of the search keyword and the selected content information. That is, for each recommendation candidate content (or content that can be provided by the device), a recommendation score (recommended suitability) is calculated based on a degree of relevance to a search keyword and/or selected content information extracted from a user's search history.

An embodiment of calculating the recommendation score is as follows.

The recommendation score calculation unit 102 is a variable (parameter) of an equation for calculating the recommendation score. (1) Whether the search keyword is included in the detailed information of the recommendation candidate content, (2) Detailed information of the recommendation candidate content Whether or not a related keyword of the search keyword is included in the, (3) a degree of relation between a category of the recommendation candidate content and the search keyword, (4) a degree of relation between the category of the selected content and the category of the recommendation candidate content, ( 5) A degree of relevance between the keyword associated with the selected content and the keyword associated with the recommended candidate content may be used. The scores calculated by the above (1) to (5) are referred to as the first score to the fifth score, respectively.

At this time, the'relevance' in (1) to (5) above may use previously analyzed keyword-category association information and/or keyword-keyword association information.

In addition, the recommended score may be calculated by putting an appropriate weight for each of the variables (first score to fifth score). For example, by placing the largest weight on (1) (for example, 2) and a smaller weight (for example, 0.5) for (2) to (5), the recommendation score of the corresponding content may be calculated. This is a method of calculating a recommendation score by placing a high proportion of the search keyword itself, which is expected to be the most intuitively relevant.

Alternatively, the highest weight (for example, 1.5) is assigned to (5), and a smaller weight (for example, 1.2) is assigned to (1) to (4), and the recommendation score of the corresponding content can be calculated. This is a method of comparing the entire keyword set of each of the selected content-candidate content and putting more weight on the potential interests of the user that are not directly revealed.

Alternatively, the recommendation score may be calculated by adding a weight determined in proportion to the number of related keywords of the search keyword included in the detailed information of the candidate content to the second score. That is, in the comparison between contents that do not directly include the search keyword, it is a method of grasping that the content containing more related keywords of the search keyword is closer to the potential interest of the user.

At this time, the related keyword of the search keyword is a related keyword whose degree of relevance to the search keyword is greater than or equal to a predetermined criterion, and the weight is determined by reflecting the number of the related keywords and the degree of relevance of each of the related keywords to the search keyword. May be. For example, even if the content includes 5 related keywords of the search keyword equally, the number of related keywords with a degree of relevance to the content greater than a predetermined criterion (e.g., 0.7) may be different, and in this case, more criteria Content including the above keywords may be determined as content suitable for recommendation. In addition, even if the number of related keywords having a degree of relevance equal to or greater than the standard is the same, if the relevance value is reflected in the weight value, the higher the relevance, the closer to the user's interest may be determined.

The recommendation score calculation unit 102 calculates the recommendation score by summing the first score to the fifth score at a predetermined reflection ratio, and the predetermined reflection ratio may be updated according to the recommendation response result of the user. have. For example, the recommendation score calculation unit 102

W1 * (first score) + W2 * (second score) + W3 * (third score) + W4 * (fourth score) + W5 * (5th score), (where W1, W2, W3, W4, W5 is the weight of each score)

The recommendation score may be calculated by the formula such as, in this case, W1 to W5 may be continuously learned by tracking and analyzing the recommendation reaction history of the corresponding user.

When the recommendation score for each candidate content is calculated as described above, the recommended content determination unit 103 of the content recommendation device 100 determines the recommended content to be recommended to the user according to the calculated recommendation score.

For example, in various ways, such as recommending the top 10 contents in order of recommendation score, recommending all contents with a recommendation score above a certain standard, or excluding the contents previously purchased by the user among the top 20 recommendation scores. Recommended content can be determined.

In this way, the content recommendation device 100 can generate a recommendation result even when there is no history of purchasing the content of the user, and by continuously learning the user's preference category and keyword from the search history, it is possible to analyze customer preferences. Accuracy can be improved.

2 is a block diagram showing a content recommendation service system according to an embodiment of the present invention.

The content recommendation service system may include a content recommendation device 100 and a user terminal 10. At this time, the content recommendation device 100 extracts one or more of the selected content information selected by the user from a search keyword and a search result according to the search keyword from the user's content search history, and one of the search keyword and the selected content information Based on the above, a recommendation score for at least one recommendation candidate content may be calculated, and recommended content to be recommended to the user may be determined according to the calculated recommendation score. In addition, the user terminal 10 may display recommended content recommended by the content recommendation device.

The user terminal 10 is a terminal that can access a service provided by the content recommendation device 100 and receive content provided from the content recommendation device 100, and may be, for example, a mobile terminal or a computer. The user terminal 10 may access a service provided by the content recommendation apparatus 100 through a series of login processes by a specific subscriber.

Accordingly, the user terminal 10 may request content recommendation for a specific category, and accordingly, may display recommended content provided from the content recommendation device 100. That is, when a specific content category (eg, movie) is input or selected by the user, the user terminal 10 may request the content recommendation device 100 to recommend content for this specific category (eg, movie).

Alternatively, the user terminal 10 has previously set recommendation request information for each time slot for content recommendation request, and when reaching a specific time zone, recommends content for a specific content category (eg, entertainment) according to the recommendation request information for the corresponding time zone. May be requested from the content recommendation device 100. In addition, various methods of requesting content recommendation for a specific content category in relation to a subscriber may be realized by the user terminal 10.

When the content recommendation device 100 provides content, that is, recommended content according to a content recommendation request, the user terminal 10 may immediately display content access information for viewing recommended content, such as recommended content, or If the display is selected, the content access information of the recommended content may be displayed.

Meanwhile, the user terminal 10 may provide content search history information to the content recommendation device 100. This may be done through the user terminal 10 or the content server. If necessary, the user terminal 10 may transmit the user's prior consent for providing the information.

The content recommendation device 100 may analyze a search history of digital content of a specific user, analyze a keyword of interest and a category of interest of the corresponding user, and recommend personalized content using the result. In particular, the server may identify the user's interest content, interest keyword, interest category, etc. by analyzing a log (e.g., search keyword, content searched among search results, etc.) . When a user's interest content, interest keyword, and the like are analyzed by the method and apparatus disclosed herein, a recommendation of the content most closely related to the interest may be provided.

In order to efficiently select recommended content according to the characteristics of the user analyzed from the search history, the content recommendation device 100 may perform a preliminary analysis on the available content (recommended candidate content) held. That is, the content recommendation device 100 may perform a task of analyzing/classifying the characteristics of all recommended candidate content so that the user's search history (eg, a specific search keyword) and the mapping (mapping) to determine the recommended content. have.

As an embodiment, the content recommendation device 100 may generate and store association information between a category-keyword, a keyword-keyword, and a keyword-category of recommendation candidate content.

In general, the digital content market includes not only information such as content ID, content name, content price, content provider, etc. for each content, but also detailed information about each content (eg, type, category, airing date, related keywords, etc.). have. Using this, the content recommendation device 100 may generate category information, related keyword information, and the like for each content from the detailed information. At this time, the category information may have a system that allows multiple levels or overlapping according to the nature of the content market. Meanwhile, the related keyword information is primarily extracted through a morpheme analyzer to extract a set of nouns, and keywords that are significant within the set can also be extracted through the TF-IDF (Term Frequency-Inverse Document Frequency) method within the category of the content. have.

In this way, the content recommendation device 100 may generate/classify information such as a content ID, a category ID, and a keyword ID of each content and store the corresponding information. According to the number of keywords extracted from the content, one or more'category-keyword' information may be provided, and a relationship between each keyword and a category, a relationship between each keyword, etc. may be determined and stored. Tables 1 and 2 described in FIG. 1 are examples of related information between a keyword-category and a keyword-keyword generated for arbitrary content.

Thereafter, the content recommendation device 100 may extract one or more of the selected content information selected by the user from a search keyword and a search result according to the search keyword from the content search history of a specific user.

That is, by analyzing the history of a user's search for content on a specific internet site (e.g., content market), (1) information on a search keyword (or search word), and (2) a user from among the search results according to the search keyword. Extracts information on the content selected and retrieved. Hereinafter, the above (2) is referred to as "selected content information". In this case, the selected content information may include an ID of the corresponding content, a category of the corresponding content, detailed information of the corresponding content, a related keyword of the corresponding content, and a degree of a relationship between the related keyword and the category.

Next, the content recommendation device 100 calculates a recommendation score for at least one recommendation candidate content based on at least one of the search keyword and the selected content information. That is, for each recommendation candidate content (or content that can be provided by the device), a recommendation score (recommended suitability) is calculated based on a degree of relevance to a search keyword and/or selected content information extracted from a user's search history.

An embodiment of calculating the recommendation score is as follows.

As a variable (parameter) of an equation for calculating the recommendation score, the content recommendation device 100 includes (1) whether the search keyword is included in the detailed information of the recommendation candidate content, and (2) the detailed information of the recommendation candidate content. Whether or not a related keyword of the search keyword is included, (3) a degree of relation between a category of the recommendation candidate content and the search keyword, (4) a degree of relation between the category of the selected content and the category of the recommendation candidate content, (5) ) It is possible to use a degree of association between a keyword associated with the selected content and a keyword associated with the recommended candidate content. The scores calculated by the above (1) to (5) are referred to as the first score to the fifth score, respectively.

At this time, the'relevance' in (1) to (5) above may use previously analyzed keyword-category association information and/or keyword-keyword association information.

In addition, the recommended score may be calculated by putting an appropriate weight for each of the variables (first score to fifth score). For example, by placing the largest weight on (1) (for example, 2) and a smaller weight (for example, 0.5) for (2) to (5), the recommendation score of the corresponding content may be calculated. This is a method of calculating a recommendation score by placing a high proportion of the search keyword itself, which is expected to be the most intuitively relevant.

Alternatively, the highest weight (for example, 1.5) is assigned to (5), and a smaller weight (for example, 1.2) is assigned to (1) to (4), and the recommendation score of the corresponding content can be calculated. This is a method of comparing the entire keyword set of each of the selected content-candidate content and putting more weight on the potential interests of the user that are not directly revealed.

Alternatively, the recommendation score may be calculated by adding a weight determined in proportion to the number of related keywords of the search keyword included in the detailed information of the candidate content to the second score. That is, in the comparison between contents that do not directly include the search keyword, it is a method of grasping that the content containing more related keywords of the search keyword is closer to the potential interest of the user.

At this time, the related keyword of the search keyword is a related keyword whose degree of relevance to the search keyword is greater than or equal to a predetermined criterion, and the weight is determined by reflecting the number of the related keywords and the degree of relevance of each of the related keywords to the search keyword. May be. For example, even if the content includes 5 related keywords of the search keyword equally, the number of related keywords with a degree of relevance to the content greater than a predetermined criterion (e.g., 0.7) may be different, and in this case, more criteria Content including the above keywords may be determined as content suitable for recommendation. In addition, even if the number of related keywords having a degree of relevance equal to or greater than the standard is the same, if the relevance value is reflected in the weight value, the higher the relevance, the closer to the user's interest may be determined.

The content recommendation device 100 calculates the recommendation score by summing the first score to the fifth score at a predetermined reflection ratio, and the predetermined reflection ratio may be updated according to the recommendation response result of the user. . For example, the content recommendation device 100

W1 * (first score) + W2 * (second score) + W3 * (third score) + W4 * (fourth score) + W5 * (5th score), (where W1, W2, W3, W4, W5 is the weight of each score)

The recommendation score may be calculated by the formula such as, in this case, W1 to W5 may be continuously learned by tracking and analyzing the recommendation reaction history of the corresponding user.

When the recommendation score for each candidate content is calculated as above, the content recommendation device 100 determines recommended content to be recommended to the user according to the calculated recommendation score.

For example, in various ways, such as recommending the top 10 contents in order of recommendation score, recommending all contents with a recommendation score above a certain standard, or excluding the contents previously purchased by the user among the top 20 recommendation scores. Recommended content can be determined.

3 is a flowchart illustrating a content recommendation method according to an embodiment of the present invention.

The content recommendation method described below may be performed by a content recommendation device.

The content recommendation device may analyze a digital content search history of a specific user, analyze a keyword of interest and an interest category of the user, and recommend personalized content using the result. In particular, the server may identify the user's interest content, interest keyword, interest category, etc. by analyzing a log (e.g., search keyword, content searched among search results, etc.) . When a user's interest content, interest keyword, and the like are analyzed by the method and apparatus disclosed herein, a recommendation of the content most closely related to the interest may be provided.

In order to efficiently select recommended content according to the characteristics of the user analyzed from the search history, the content recommendation device may perform a preliminary analysis on the available content (recommended candidate content) held. That is, the content recommendation device may perform a task of analyzing/classifying the characteristics of all recommended candidate contents so that the user's search history (eg, a specific search keyword) may be mapped to determine the recommended content.

As an embodiment, the content recommendation device may generate and store association information between a category-keyword, a keyword-keyword, and a keyword-category of recommended candidate content.

In general, the digital content market includes not only information such as content ID, content name, content price, content provider, etc. for each content, but also detailed information about each content (eg, type, category, airing date, related keywords, etc.). have. Using this, the content recommendation device may generate category information, related keyword information, and the like for each content from the detailed information. At this time, the category information may have a system that allows multiple levels or overlapping according to the nature of the content market. On the other hand, related keyword information is primarily extracted through a morpheme analyzer to extract a set of nouns, and keywords that are significant within the set can also be extracted through the TF-IDF (Term Frequency-Inverse Document Frequency) method within the category of the content. have.

In this way, the content recommendation device may generate/classify information such as a content ID, a category ID, and a keyword ID of each content and store the corresponding information. According to the number of keywords extracted from the content, one or more'category-keyword' information may be provided, and a relationship between each keyword and a category, a relationship between each keyword, etc. may be determined and stored. Tables 1 and 2 described in FIG. 1 are examples of related information between a keyword-category and a keyword-keyword generated for arbitrary content.

Thereafter, the content recommendation device may extract one or more of a search keyword and selected content information selected by the user from among a search keyword and a search result according to the search keyword from a content search history of a specific user.

That is, by analyzing the history of a user's search for content on a specific internet site (e.g., content market), (1) information on a search keyword (or search word), and (2) a user from among the search results according to the search keyword. Extracts information on the content selected and retrieved. Hereinafter, the above (2) is referred to as "selected content information". In this case, the selected content information may include an ID of the corresponding content, a category of the corresponding content, detailed information of the corresponding content, a related keyword of the corresponding content, and a degree of a relationship between the related keyword and the category.

Next, the content recommendation device calculates a recommendation score for at least one recommendation candidate content based on at least one of the search keyword and the selected content information. That is, for each recommendation candidate content (or content that can be provided by the device), a recommendation score (recommended suitability) is calculated based on a degree of relevance to a search keyword and/or selected content information extracted from a user's search history.

An embodiment of calculating the recommendation score is as follows.

The content recommendation device is a variable (parameter) of an equation for calculating the recommendation score, including (1) whether the search keyword is included in detailed information of the recommendation candidate content, and (2) the search keyword in detailed information of the recommendation candidate content. Whether or not a related keyword of is included, (3) the degree of relation between the category of the recommendation candidate content and the search keyword, (4) the degree of relation between the category of the selected content and the category of the recommended candidate content, (5) the selected content It is possible to use a degree of relevance between a related keyword of and a related keyword of the recommended candidate content. The scores calculated by the above (1) to (5) are referred to as the first score to the fifth score, respectively.

At this time, the'relevance' in (1) to (5) above may use previously analyzed keyword-category association information and/or keyword-keyword association information.

In addition, the recommended score may be calculated by putting an appropriate weight for each of the variables (first score to fifth score). For example, by placing the largest weight on (1) (for example, 2) and a smaller weight (for example, 0.5) for (2) to (5), the recommendation score of the corresponding content may be calculated. This is a method of calculating a recommendation score by placing a high proportion of the search keyword itself, which is expected to be the most intuitively relevant.

Alternatively, the highest weight (for example, 1.5) is assigned to (5), and a smaller weight (for example, 1.2) is assigned to (1) to (4), and the recommendation score of the corresponding content can be calculated. This is a method of comparing the entire keyword set of each of the selected content-candidate content and putting more weight on the potential interests of the user that are not directly revealed.

Alternatively, the recommendation score may be calculated by adding a weight determined in proportion to the number of related keywords of the search keyword included in the detailed information of the candidate content to the second score. That is, in the comparison between contents that do not directly include the search keyword, it is a method of grasping that the content containing more related keywords of the search keyword is closer to the potential interest of the user.

In this case, the related keyword of the search keyword is a related keyword whose degree of relevance to the search keyword is greater than or equal to a predetermined criterion, and the weight is determined by reflecting the number of the related keywords and the degree of relevance of each of the related keywords to the search keyword. May be. For example, even if the content includes 5 related keywords of the search keyword equally, the number of related keywords with a degree of relevance to the content greater than a predetermined criterion (e.g., 0.7) may be different, and in this case, more criteria Content including the above keywords may be determined as content suitable for recommendation. In addition, even if the number of related keywords having a degree of relevance equal to or greater than the standard is the same, if the relevance number is reflected in the weight value, the higher the relevance, the closer to the user's interest may be determined.

The content recommendation device calculates the recommendation score by summing the first score to the fifth score at a predetermined reflection ratio, and the predetermined reflection ratio may be updated according to a recommendation response result of the user. For example, the content recommendation device

W1 * (first score) + W2 * (second score) + W3 * (third score) + W4 * (fourth score) + W5 * (5th score), (where W1, W2, W3, W4, W5 is the weight of each score)

The recommendation score may be calculated by the formula such as, in this case, W1 to W5 may be continuously learned by tracking and analyzing the recommendation reaction history of the corresponding user.

When the recommendation score for each candidate content is calculated as above, the content recommendation device determines the recommended content to be recommended to the user according to the calculated recommendation score.

For example, in various ways, such as recommending the top 10 contents in order of recommendation score, recommending all contents with a recommendation score above a certain standard, or excluding the contents previously purchased by the user among the top 20 recommendation scores. Recommended content can be determined.

The embodiments of the present invention have been described above, but those of ordinary skill in the relevant technical field can add, change, delete or add components within the scope not departing from the spirit of the present invention described in the claims. Accordingly, the present invention can be variously modified and changed, and it will be said that this is also included within the scope of the present invention.

100: content recommendation device
101: search history analysis unit
102: recommendation score calculation unit
103: Recommended content determination unit

Claims (13)

A search history analysis unit for extracting a category and a related keyword of the selected content selected by the user from a search keyword and a search result according to the search keyword from a user's content search history;
A recommendation score calculation unit for calculating a recommendation score based on a relationship between the search keyword and the category of the selected content and a related keyword for each recommendation candidate content;
And a recommended content determination unit configured to determine recommended content to be recommended to the user from among the recommended candidate content according to the calculated recommendation score,
The recommended score calculation unit,
The recommendation is made by assigning a weight determined in proportion to the number of related keywords of the search keyword included in the detailed information of the recommendation candidate content to a score according to whether the detailed information of the recommended candidate content includes a related keyword of the search keyword. Content recommendation device comprising calculating a score. delete delete delete delete delete delete delete delete delete delete delete delete

Images