KR100925511B1 - Method and system for classifing content using collaborative filtering - Google Patents

Abstract

The present invention relates to a content classification method and system for classifying unclassified content without a classification name by using previously classified content having a classification name according to a classification system.

Specifically, the present invention provides a content classification method using collaborative filtering, comprising: (a) transmitting content usage information including content information, user information, and usage history information; (b) dividing the content usage information into pre-classified content usage information and non-classified content usage information according to whether a classification name exists according to a classification system; (c) calculating preferences for each user's classification name according to the usage history information from the previously classified content usage information; (d) calculating preferences of respective users based on the usage history information from the unclassified contents usage information; (e) calculating a similarity between the classification name and the content based on the preference by classification name and the preference by content; (f) generating classification name-content group information according to the similarity; (g) providing a content classification method and system using cooperative filtering, including classifying the unclassified content into a corresponding classification name according to the classification name-content group information.

Description

Method and system for classifing content using collaborative filtering}

The present invention relates to a content classification method and system for classifying unclassified content without a classification name by using previously classified content having a classification name according to a classification system.

More specifically, in order to classify unclassified content having no classification name according to the classification system, the present invention calculates the preference of each user for the classification name and the content preference of each user for the unclassified content. The present invention relates to a content classification method and system using collaborative filtering for classifying unclassified content into a corresponding classification name by generating classification name-content group information using the classification name and the similarity between the contents.

Recently, as society has entered the age of full-scale informatization, the development of the Internet and the development of information and communication technologies have been rapidly developed, and the electronic commerce based on the Internet is rapidly spreading.

An example of such e-commerce is a business-to-consumer electronic commerce called B2C (Business-to-Consumer), which refers to all types of electronic transactions between suppliers and consumers via the Internet. At this time, the general object of the electronic commerce can be any of the tangible and intangible, which is referred to as content (hereinafter).

On the other hand, in general e-commerce, the consumer is often the unspecified number of customers, the supplier needs to establish and operate a more efficient marketing strategy. As a result, many suppliers are actively implementing 'personalization service' that provides valuable services to the users by actively using customer information related to transactions. At this time, the 'personalization service' refers to a service that selectively selects data that is expected to be of value to the consumer by using information implicitly or explicitly provided by the user, such as user's personal information, preference, or consumption form.

A representative example of such personalization service is 'content recommendation information service through collaborative filtering', where 'collaborative filtering' predicts a user's interest based on taste information obtained from the user. A method of selecting a content suitable for a specific user by using similarity between the user and / or the content, on the premise that a user having similar tendency prefers similar content, and a content recommendation information service is a collaboration method. Refers to a service of recommending and introducing information on specific content obtained through filtering to a corresponding user.

Accordingly, general collaborative filtering may be classified into 'user to user collaborative filtering' using similarity between users and 'content to content collaborative filtering' using similarity between contents.

In detail, 'collaboration filtering between users' groups information that is similar to each other by using information obtained from users, and then recommends content to a specific user by using contents purchase information or usage information of other users in the group. On the other hand, 'collaboration filtering between contents' is a form of grouping contents that are often used together within a certain time period, and when a user uses a specific content, other content recommendation information in the group is typically provided. .

That is, in the case where User 1 uses Content 1 while User 1 and User 2 are grouped, a form of recommending Content 1 to User 2 may be a good example of 'collaboration filtering between users'. When User 1 uses Content 1 in a grouped state, a form of providing recommendation information for Content 2 may be a good example of 'collaboration filtering between contents'. Recently, a new type of user-content collaborative filtering method that combines 'collaboration filtering between users' and 'collaboration filtering between contents' has been introduced to provide more efficient content recommendation information service.

On the other hand, given the recent trend in which the type of content is rapidly increasing, the clear classification of the content has an important effect on both the content provider and the consumer. In this case, 'classification of content' refers to grouping or classifying contents according to a predetermined classification system. In the case of lack of clarity of content classification, not only accurate content selection but also duplicate classification or classification The possibility of omission is large.

Accordingly, a method of finding and organizing clues of classification from the inside of a content is generally used, but this is possible only when a context is easily extracted because of words and symbols, such as a document. Substantial classification is impossible.

Therefore, a specific strategy for automatically and clearly classifying content is required.

Accordingly, an object of the present invention is to provide a content classification method that can automatically and quickly classify content using collaborative filtering.

That is, the present invention, when it is difficult or impossible to find the clue of the classification from the inside of the content, the content of each user for each user's content preferences and unclassified content of each user with respect to the existing classification content with the classification name The object of the present invention is to provide a specific method for automatically classifying unclassified contents into a corresponding classification name by determining similarity between classification names and contents through star preference, and generating classification name-content group information using the similarity.

In order to achieve the above object, the present invention provides a content classification method using collaborative filtering, comprising: (a) transmitting content usage information including content information, user information, and usage history information; (b) dividing the content usage information into pre-classified content usage information and non-classified content usage information according to whether a classification name exists according to a classification system; (c) calculating preferences for each user's classification name according to the usage history information from the previously classified content usage information; (d) calculating preferences of respective users based on the usage history information from the unclassified contents usage information; (e) calculating a similarity between the classification name and the content based on the preference by classification name and the preference by content; (f) generating classification name-content group information according to the similarity; (g) providing a content classification method using collaborative filtering comprising the step of classifying the unclassified content into a corresponding classification name according to the classification name-content group information.

In this case, after the step (b) and before the step (c), the content information of the pre-categorized content use information is further replaced by a corresponding classification name. Wherein the usage history information includes usage type information of the contents, usage time information of the contents, and usage frequency information of the contents, wherein step (c) and / or the Step (d) is a step of calculating a preference for each type of use according to the usage type information, a step of calculating a preference for each use time according to the use time information, and a step of calculating a preference for each use number according to the use frequency information. Including at least one of, the preference by classification name and / or the content preference is the preference by the type of use, preference by the time of use, the It is characterized in that it is calculated through at least one of the number of preferences.

In addition, the preference for each type of use is calculated as a weight that is pre-allocated for each type of use of the use type information, the preference for each use time, the specific decrease according to the interval between the latest time and the current time of the use time information It is characterized in that it is calculated with reference to the relationship, characterized in that the specific reduction relationship is in accordance with the sigmoid attenuation curve.

In addition, the preference for each use frequency is calculated by referring to a specific increase relation according to the use frequency of the use frequency information, and after the step (c) before the step (e), the predetermined abnormal user determination criteria And filtering the user's preferences according to the classification names according to the user's preferences. The method may further include filtering.

In addition, the abnormal user judgment criteria, the occupation of the user is associated with the creation, development, processing, sale or advertising of the content, the nationality of the user is out of the country, or the user within the minimum usage time interval for the same content It is characterized in that at least one of using more than the maximum number of times, or the user performs the same behavior more than the maximum number of times within the minimum use time interval, step (e), (e1) through the preference by the classification name Generating a preferred user group for each classification name; (e2) calculating the similarity between the classification name and the content based on the content-specific preference of the user group, wherein the user information includes at least one user of the gender, age, and region of the user. In addition to the attribute information, the preferred user group is generated in consideration of the user attribute.

The method may further include filtering the corresponding content when the similarity is less than the reference value after comparing the similarity with the reference value after the step (e) and before the (f) step.

In addition, the present invention provides a content classification method using collaborative filtering, comprising the steps of: (a) collecting pre-classified content usage information and unclassified content usage information having information on a classification name; (b) calculating a preference of each user from the pre-classified content usage information to generate a predetermined user group; (c) selecting predetermined unclassified contents preferred by the user group from the unclassified contents usage information; (d) providing a content classification method using collaborative filtering, wherein the classification name of the selected non-classified content is set as a classification name of pre-classified content preferred by the user group.

In this case, the content usage information includes content information, user information, usage history information, and the user history information includes at least one of usage type information of the content, usage time information of the content, and usage count information of the content. The step (b) may include generating the user group whose preference for the existing classified content exceeds a predetermined criterion in the existing classified content using information having the same classification name. It further comprises.

In addition, the present invention provides a content providing server interface for receiving content use information containing content information, user information, usage history information; A preference calculation unit for each classification name that selects previously classified content usage information having a classification name according to a classification system among the contents usage information and calculates a preference for each classification name of each user; A content preference calculator for selecting unclassified content usage information without a classification name according to a classification system from the content usage information to calculate a preference for each user; A similarity calculation unit configured to calculate a similarity between the classification name and contents based on the preference by classification name and the preference by content, and generate classification name-content group information based on the similarity; Provided is a content classification system using collaborative filtering including a classification unit for classifying the unclassified content into a corresponding classification name by using the classification name-content group information.

In this case, the similarity calculating unit may generate a preferred user group for each classification name based on the preference for each classification name, and calculate the similarity between the classification name and contents based on the preference for each content of the preferred user group. A content usage information database in which the usage information is stored and managed; A preference information database for each classification name in which preferences for each classification name are stored and managed; A content preference information database for storing and managing the content preferences; And the classification name-content group database in which the unclassified classification name-content group information is stored.

As described above, the present invention is to determine the similarity between the classification name and the content by using each user's classification name preference for the previously classified content with the classification name and each user's content preference for unclassified content, Unclassified content is automatically classified into a corresponding classification name by generating classification name-content group information using the similarity.

As a result, the present invention can classify the content directly reflecting the user's preference, so that the reliability is high and more accurate content recommendation information can be selected and provided.

Hereinafter, the present invention will be described in detail with reference to the accompanying drawings.

Before the full description, several terms to be used in the present specification are summarized, and the term 'content' is an object provided to a user by a supplier. It may be a data file such as a file, a video file, a document file, etc. A 'classification name' is a classification criterion for contents according to an arbitrary classification system. In the case of a category or music file to which the article or content belongs, it may be a genre. have. Therefore, each content is classified under a classification name, and the absence of a classification name means that it is impossible to know which category it belongs to even by referring to internal attributes of the corresponding content. In addition, the use of content is a concept that encompasses all forms such as not only purchasing content, but also inquiry, introduction, scrap to the personal area, and a short testimonial, such that the definition of the term is consistently used herein. Will be used in the same sense.

1 is a block diagram showing a preferred aspect of a content classification system (hereinafter, simply referred to as a content classification system) using collaborative filtering according to the present invention.

As shown, the content classification system according to the present invention includes a content providing server interface unit 110, a content usage information storage unit 112, a content usage information management DB 114, including a content usage information management DB 114 and Preference information management DB 124 by classification name, Preference calculation unit 126 by content, Preference information management DB 128 by content, Similarity calculation unit 130, Classification name-content information management DB 132, and classification name And a content information providing unit 134.

Looking at each of them in detail as follows.

First, the content providing server interface 110 is a part interfacing with a separate content providing server (not shown).

The content providing server may provide a user with personalized content recommendation information and provide corresponding content according to a user's selection. Accordingly, the content providing server generates and manages content usage information including user information on the content user, content information on the corresponding content, and usage history information on the user's use of the content, and the content usage information according to the present invention. The content providing server interface 110 of the content classification system is transmitted. In this case, the content usage information may be transmitted to the content providing server interface 110 for each predetermined time.

Next, the content usage information storage unit 112 stores the content usage information received through the content providing server interface 110 in the content usage information management DB 114. In this case, if necessary, the content usage information storage unit 112 may convert or process the content usage information into an appropriate format.

Next, the content usage information management DB 114 stores and manages content usage information delivered through the content usage information storage 112. In this case, the content usage information management DB 114 may store the content usage information by user, by content, and by usage history. An example thereof is illustrated in FIG. 2.

That is, FIG. 2 is an exemplary view of a content usage information table stored and managed in the content usage information management DB 114, and includes user information for user identification, content information for content identification, and use of content. History information is included. In this case, the usage history information may include at least one of usage type information, usage count information, and usage time information. Although not shown in the drawing, the user information includes user attribute information such as gender, age, region, and occupation of the user. The content may be classified into pre-classified content having a classification name and non-classified content having no classification name according to the presence or absence of a classification name.

1 again, the content usage information management DB 114 divides the content usage information into pre-classified content usage information and unclassified content usage information according to the presence or absence of the classification name, and the pre-classified content usage information is classified by the preference name calculation unit ( On the other hand, the unclassified content usage information is transmitted to the content preference calculator 126.

Next, the classification preference preference unit 122 calculates the preference for each classification name by using the classification content usage information. To this end, the classification preference unit 122 analyzes each user's usage history and calculates at least one of preferences by type of use, preferences by use time, and preferences by frequency of use based on the classification name, and based on these, the classification name of each user. Calculate your preferences.

Attached FIG. 3 is an exemplary diagram of a pre-classified content usage information table by the category-specific preference calculator 122, and compared to FIG. 2, it can be seen that each piece of content information is replaced with a corresponding category name. Accordingly, the classification preference unit 122 calculates the preference by type, the preference by use time, and the preference by number of times based on the usage history information of each user, and calculates the preference by classification name of each user based on these.

At this time, the preference for each type of use can be calculated using the weights pre-assigned for each type of use.In the figure, for convenience, the types of use are divided into three categories: appreciation, purchase, and inquiry. Each type of use is assigned a weight of 1, 5, 3. Therefore, since the user 1 indicates the usage history of appreciation, purchase, and inquiry for the classification names-ballads of numbers 1 and 5, the preference for each type of use may be scored to 18.

In addition, the preference for each use time can be calculated by referring to a specific decrease relation according to the interval between the recent use time and the present time among the use time of each user's classification name. In this case, the reason for considering the use time is the preference over time The reduction is taken into account, and the specific reduction relationship for improving reliability can be based on the sigmoidal decay curve of the S-shape which is common in natural decay. Accordingly, the user's preference for each use time may be scored as 5 on the premise that the use time of the number 5 is more recent.

In addition, the preference by frequency can be calculated by referring to a specific increase relation according to the number of times of classification by each user. In general, although the preference of a user who used n times is relatively higher than that of a user who has used a specific name once, Considering personal tendencies, it is not necessarily quantified by n times. Therefore, the specific increase relation for improving reliability may be in accordance with the law of diminishing marginal utility, and thus the user 1 preference may be scored as 2.

As a result, the preference of each classification name for the classification name 'ballard' of the user 1 may be scored as 25.

At this time, the method of calculating the preference by classification name, including the preference by type of use, the preference by time of use, and the preference by number of times of use may be modified as necessary. Therefore, the above description is merely one example. It will be true.

In addition, the classification preference preference unit 122 may increase the reliability by filtering the preference of the classification name of a specific user according to a predetermined abnormal user determination criteria. This is to reduce the likelihood of manipulation of the preference by classification name due to unintended user's intentional behavior. For this purpose, the preference calculation unit 122 compares the user's attribute information and usage history information with preset abnormal user judgment criteria. Filter user preference by category name.

At this time, among the criteria for abnormal user's judgment, the criteria according to the user's attribute information include when the user's job is related to the creation, development, processing, sale, or advertisement of the corresponding content, or where the user's living area is outside the country of sale of the corresponding content. The criterion according to the usage history information may be the case where the same content is accessed more than the maximum number of times within the minimum access interval time, or a specific user repeats the same behavior more than the maximum number of times within the minimum execution time interval. .

Among these, the criterion according to the user attribute information can be easily understood even if there is no separate explanation, and thus, the criterion based on the usage history information is briefly described. For example, the use history information for the user name 2 'ballard' is classified. If the time interval of 'yy / mm / dd hh: mm: ss 6_1 ~ yy / mm / dd hh: mm: ss 6_3' is within the preset minimum access time interval and the maximum number of times of usage is 3, User 2 Actions corresponding to the numbers 6 to 8 in the case of "access to the same content more than the maximum number of times of access within the minimum access interval time". Therefore, the classification preference unit 122 determines that a corresponding action of the user 2 is a filtering target, and excludes the preference by classification name.

In addition, the preference calculator 122 by the classification name may separate the preference by classification name so that the characteristics of the users belonging to the minority group are not ignored, and then calculate the preference by classification name according to each user's attributes. To this end, the preference calculator 122 by classification name divides the preferences of each user's classification name shown in FIG. 2 according to user attributes and merges them for each same attribute.

That is, FIGS. 4 and 5 are table views showing preferences according to classification names according to user attributes, respectively. FIG. 4 shows preferences by classification names of women in Seoul / Gyeonggi area in consideration of region and gender in user attribute information. 5 represents the preference by classification name of males in Chungcheong area.

In addition, the preference by classification name of each user described with reference to FIG. 2 and the preference by classification name according to the user attributes described with reference to FIGS. 4 and 5 are stored in the preference information management DB 124 by classification name.

Meanwhile, the content preference calculator 126 calculates the content preference of each user by using the unclassified content usage information. To this end, the content preference calculator 126 analyzes the usage history of each user and calculates at least one of preferences by type of use, preferences by use time, and preferences by number of times based on the content, and calculates preferences based on the content. do.

6 is a diagram illustrating an unclassified content usage information table by the content preference calculator 126, which is different from that in FIG. 2 in that a preference for each content is calculated rather than a classification name. The same may apply. As a result, the preference by content type 2 for user 1 of number 3 may be scored as 6, the preference by time of use is 3, the preference by frequency of use is 1, and the preference for content 2 may be scored as 10.

In addition, the content preference calculator 126 may increase the reliability by filtering the content preferences of a specific user according to a predetermined abnormal user judgment criterion, and the content according to the user attribute so that the characteristics of the users belonging to the minority group are not ignored. After classifying the preferences, merge them by the same property.

Accordingly, FIG. 7 illustrates preferences of women in Seoul / Gyeonggi area by content in consideration of region and gender in user attribute information, and FIG. 8 illustrates preferences of contents of men in Chungcheong area. The detailed operation of the preference calculator 126 for each content may be easily understood by referring to the description of the preference calculator 122 for each category, as described above.

In addition, the preferences for each user according to the contents described in FIG. 6 and the preferences for each content according to the user attributes described in FIGS. 7 and 8 are stored in the content preference information management DB 128.

Referring back to FIG. 1, the similarity calculating unit 130 is stored in the classification preference and content preference information management DB 128 according to the classification name preference and the user attribute of each user stored in the classification information preference information management DB 124. Calculate the similarity between 'classification name' and 'content' by using each user's content preference and user's preference according to user attributes, and group the classification name and content with high similarity to 'classification-content group information' Create In this case, the classification name-content group information may correspond to a category name and contents that are correlated with each other in a matrix form or correspond to one-to-one, one-to-one, and plural-to-one.

More specifically, the similarity calculation process between the classification name and the content by the similarity calculator 130 will be described.

First, the similarity calculator 130 generates a preferred user group for each classification name by using the preference for each classification name of each user stored in the preference information management DB 120 for each classification name. Here, the preferred user group for each classification name may be composed of people whose preference for the same classification name is greater than or equal to a preset threshold. Next, the similarity calculator 130 selects the contents used by the users in the preferred user group to calculate the similarity between the classification name and the corresponding content, and generates similar classification name-content group information based on the classification name and the content when the similarity is higher than the reference value.

For the sake of understanding, the contents A, B, C, and D are contents with classification names, which are classified as ballads, and contents E, F, G, H, L, M, N, X, and Y. A, Z is the content A, B, C, D, E, F, G, H, and the user A, B, C, When D, E, X, Y, and Z are used, and the user bottle uses contents A, B, C, D, E, L, M, and N, the similarity calculation unit 130 performs classification for each user's classification name. The preferences group user packs, packs, bottles, and tablets into groups of preferred users with the classification name Ballad.

Next, the similarity calculator 130 selects the unclassified contents used by the user, group, and bottle in the preferred user group of the classification name 'ballad', and calculates the similarity between the selected contents and the classification name, and the preferred user of the classification name 'ballard'. Since all users in the group used content E, the similarity between classification name 'ballard' and content E was 100%, and only one of the users in the preferred user group of category name 'ballard' used content F. The similarity between F is about 33%, and similarly, the similarity between the classification name 'ballard' and the contents G, H, L, M, N, X, Y, Z is about 33%.

In this case, assuming that the reference value for the predetermined similarity determination is 50%, since the similarity between the classification name 'ballard' and the content E is greater than the reference value, the classification name-content information group for 'ballard'-'content E' is generated. Meanwhile, a known clustering method may be used to calculate the similarity, and specific examples thereof may include algorithms used in common collaborative filtering such as Pearson correlation, popularity difference method, and cosine similarity method. A detailed clustering method for the similarity calculation will not be described separately.

In this case, in order to further increase the accuracy of the classification name-content group information, it is preferable to consider the preferences by classification name and content preference according to user attributes. When creating a preferred user group by classification name, the classification name according to each user attribute is preferred. Preference may be considered. That is, in the above example, the user pack, group, bottle, and tablet may be women who reside in Seoul / Gyeonggi area arbitrarily according to user attributes, or may be a preferred user group considering other user attributes.

In addition, some weights may be considered in the generation of the classification name-content group information. For example, when the use interval of the classification name-content used in a user preference group for each classification name exceeds a predetermined time, Inversely, time weights that are inversely applied and the same weight or when used within a predetermined time can be given high weights to increase their reliability.

In addition, the similarity calculator 130 may filter the corresponding preferred user group when the number of users in the preferred user group for each classification name is less than a preset number of users, thereby increasing the accuracy of the classification name-content group.

Next, the preferred user group for each classification name and the classification name-content group information generated by the similarity calculating unit 130 are stored in the classification name-content information management DB 132, and the dual classification name-content group information is classified name-content information providing unit ( 134).

Finally, the classification name-content information providing unit 134 classifies the unclassified contents into the corresponding classification names, and provides the unclassified contents with the corresponding classification names by providing them to a separate content providing server through the content providing server interface 110. do.

FIG. 9 is a flowchart illustrating a content classification method according to the present invention. The description of the present invention will be described with reference to FIG. 1.

For the content classification method according to the present invention, first, content usage information is transmitted to the content providing server interface 110. (st1) In this case, the content usage information includes user identification information as well as user gender, age, region, occupation, etc. User information containing the user attribute information of the content, content information for content identification, and usage history information for the use of the content.

Subsequently, the content usage information is stored in the content usage information management DB 114 by the content usage information storage unit 112, and the classified content usage information and the unclassified content usage information are classified according to the classification name presence (st2).

Subsequently, the preference calculation unit 122 for each classification name replaces the content identification information in the previously classified content usage information with the classification name, and calculates the preference for each type of use, time of use, and number of times of use based on usage history information of each user. Through these calculations, preferences are calculated by classification name, the preferences of specific user's classification names are filtered according to the preset abnormal user judgment criteria, and the preferences according to classification names are classified according to user attributes and merged by the same attributes (st11 ~ st15). And the preference by classification name according to the classification name and user attribute of each user is stored in the classification information preference information management DB (124).

In addition, the content preference calculator 126 calculates a preference by use type, time of use, and frequency of use based on usage history information for each user in unclassified content use information by content, and calculates a preference by content through these. It filters the preferences by specific user's classification name based on the preset abnormal user judgment criteria, classifies the content preferences according to the user attributes, and merges them by the same attributes (st21 ~ st24). Content-specific preferences based on user attributes are stored in content-specific preference information management DB 128.

Subsequently, the similarity calculator 130 generates a preferred user group for each classification name using the preference for each classification name, and generates classification name-content group information by calculating the similarity between the classification name and the content based on the content preference of the corresponding preferred user group. On the other hand, highly reliable classification name-content group information is generated by filtering content indicating a similarity below the threshold and filtering the preferred user group by a few users. (St31 to st33)

Accordingly, when the classification name-content group information is completed, the classification name-content group information is stored in the classification name-content information management DB 132, and the classification name-content information providing unit 134 performs unclassified contents according to the classification name-content group information. Classify (st34)

The above has been illustrated and described with respect to the preferred embodiment of the present invention, the present invention is not limited to this, that is, the present invention can be modified in any way according to the specific manner, which should all fall within the scope of the present invention. The scope of the present invention is more clearly shown in the following claims.

1 is a block diagram of a content classification system in accordance with the present invention.

2 is an exemplary table of content usage information according to the present invention.

3 is a diagram illustrating a preference table for each classification name of pre-classified content usage information according to the present invention.

4 and 5 are table examples of the preference by classification name according to the user attribute of the pre-classified content usage information according to the present invention, respectively.

6 is a diagram illustrating a preference table for each content of unclassified content usage information according to the present invention.

7 and 8 are table examples of the preferences for each content according to user attributes of unclassified content usage information according to the present invention.

9 is a flowchart of a content classification method according to the present invention;

Claims (20)

Content information on the pre-classified content having the classification name according to the classification system and the non-classified content without the classification name according to the classification system, the user information about the user of the content, and the type of use of the user for the content. A content classification method using collaborative filtering of a content classification system for classifying the unclassified content by using content usage information including at least one usage history information of a usage time and a usage frequency. (a) transmitting the content usage information from a content providing server to the content classification system; (b) dividing the content usage information into content classification information for the pre-classified content and the unclassified content by the content classification system; (c) calculating, by the content classification system, preferences for each classification name of the user according to the usage history information on the previously classified content; (d) After the content classification system generates a preferred user group for each classification name for the users whose preference level is greater than or equal to the reference value, the unclassified content selected by the user of the classification group preferred user group is selected, clustering by collaborative filtering is performed. Calculating a similarity between the classification name and the unclassified content; (e) generating, by the content classification system, classification name-content group information according to the classification name for the unclassified content whose similarity is equal to or greater than a reference value; And and (f) classifying the unclassified content into a corresponding classification name according to the classification name-content group information by the content classification system. delete delete The method according to claim 1, The step (c) is a step of calculating a preference for each use type in proportion to the sum of weights pre-allocated according to the use type, calculating a preference for each use time in proportion to the use time, using in proportion to the use frequency. The method of claim 1, further comprising calculating at least one of the number of preferences, wherein the classification preference is calculated based on a sum of at least one of the preference by type, usage time, and preference. delete  The method according to claim 4, The content preference method according to the use time is a content classification method using collaborative filtering reflecting a specific reduction relationship according to the interval between the latest time and the current time. The method according to claim 6, The specific reduction relationship is a content classification method using collaborative filtering according to the sigmoid attenuation curve. The method according to claim 4, The content preference method according to the number of times of use is a content classification method using collaborative filtering reflecting a specific increase relation according to the number of times of use.  The method according to claim 8, The specific increase relationship is a content classification method using collaborative filtering according to the law of diminishing marginal utility. delete The method according to claim 1, After the step (c) and before the step (d), the method further comprises the step of filtering the preferences of the user by the classification name in accordance with a predetermined abnormal user determination criteria, wherein the abnormal user determination criteria is the content of the user's job In connection with the creation, development, processing, sale or advertisement of the user, or outside the nationality of the user, the user uses the same content more than the maximum number of times within the minimum usage time interval, or the user maximum within the minimum usage time interval Content classification method using collaborative filtering which is at least one of performing the same behavior more than the number of times of use. delete The method according to claim 1, The user information further includes at least one user attribute of the gender, age, and region of the user, wherein the preferred user group for each classification name of step (d) is generated for each user attribute. delete delete delete delete Connected to the content providing server, the content information for the pre-classified content having the classification name according to the classification system and the non-classified content without the classification name according to the classification system, the user information for the user of the content, and the content A content providing server interface unit configured to receive content usage information including at least one usage history information of the user's usage type, usage time, and usage frequency of the user; A classification name preference calculator configured to select content use information of the previously classified content having a classification name according to a classification system among the content use information, and calculate a preference for each classification name of the user according to the usage history information; Generate a preferred user group for users whose preference level is higher than the reference value, select unclassified contents selected by a user of the preferred user group, and calculate similarity between the classified name and the unclassified contents through clustering by collaborative filtering; A similarity calculator configured to generate classification name-content group information according to the classification name for the unclassified content having a similarity level or more; And And a classification unit which classifies the unclassified contents into a corresponding classification name by using the classification name-content group information. The method according to claim 18, The preference calculator for each classification name calculates a preference for each use type in proportion to the sum of weights pre-assigned according to the usage type, calculates a preference for each use time in proportion to the use time, or a preference for each use frequency in proportion to the use frequency. The content classification system using the cooperative filtering, characterized in that the calculated by the classification name is calculated by the sum of at least one of the preference by the type of use, preference by use time, preference by the number of times of use. The method according to claim 18, A content usage information database in which the content usage information is stored and managed; A preference information database for each classification name in which preferences for each classification name are stored and managed; And And a classification name-content group database storing the unclassified classification name-content group information.

Images