KR101516329B1 - System and method for recommending group in social network environments - Google Patents

Abstract

The present invention relates to a system and a method for recommending a group in a social network environment. The system updates and generates activity profile information of a user, generates preference by cooperatively filtering the activity profile information through similar user search and expert search, and recommends a group by comparing the preference with the group similarity. According to the present invention, profile information is updated by collecting the recent group activity information, so the method may solve a problem caused by an existing static profile which does not reflect the recency in the changing trend of a user. Also, various and accurate group recommendation services are provided with the user in comparison with the existing method by considering profile information of other users having the similar preference with the user in the corresponding field.

Description

[0001] SYSTEM AND METHOD FOR RECOMMENDING GROUP IN SOCIAL NETWORK ENVIRONMENTS [0002]

The present invention relates to group recommendation techniques in a social network environment, and more particularly, to a group recommendation system and method in a social network environment that recommends an appropriate group to a user by using profile and collaborative filtering in a social network.

As user demands for networking, information sharing, etc. are soaring, the use of SNS (Social Network Service) is being activated. In addition, a lot of contents and services are overflowing, and a user is highly required to receive a desired content at a desired point in time. In other words, a system for appropriately recommending a content and a product, which are determined to be preferred by a user, has been widely used, and various personalization services are being made utilizing such a recommendation system. Therefore, in order to quickly reflect the various requirements of the users, researches on recommendation techniques considering user 's tendency are actively conducted. Among these recommendation methods, group recommendation is to analyze the activity or tendency of the user to provide additional groups of interest to the user or to provide specific items within the group.

In social networks, groups are classified into static groups and dynamic groups. Static groups are groups gathered for a specific purpose, and content types that are handled by groups are determined on the surface. For example, tennis clubs, movie clubs, and so on, are groups of users subscribed for special purposes, and the types of information that are shared among groups are defined. A dynamic group is a group that is incompatible with a static group and refers to a group that is temporarily generated according to a specific event or situation and then disappears. Users watching the same TV channel, and users gathered in a specific area. Recommendations for these groups can improve service satisfaction and share improved information by recommending groups that fit the user's preferences or recommending items appropriate to group members.

The general group recommendation method derives user 's tendency based on the user' s profile and provides appropriate items to the interested group or group based on the derived information. "L. Quijano-Sanchez, J. A. Recio-Garc┱a, and B. Diaz-Agudo, "Personality and Social Trust in Group Recommendations", Proc. In this paper, we propose a new method of analyzing a user's tendency by using the Thomas-Kilmann Conflict Mode Instrument (TKI) [1] in the International Conference on Tools with Artificial Intelligence, Vol.3, pp.121-126, Using this user tendency, we recommend a group containing users with similar tendencies. "L. Boratto, S. Carta, A. Chessa, M. Agelli, and M. L. Clemente, "Group Recommendation with Automatic Identification of Users Communities", Proc. In the International Workshop on Intelligent Web Interaction, Vol.3, pp.547-550, 2009, we proposed a method for grouping users with similar tendencies and recommending reliable items to group members [2]. "M. Gartrell, X. Xing, Q. Lv, A. Beach, R. Han, S. Mishra, and K. Seada, "Enhancing Group Recommendation by Incorporating Social Relationship Interactions", Proc. International ACM SIGGROUP Conference on Supporting Group Work, pp. 97-106, 2010] proposed a recommendation company that provides items to groups using social characteristics [3]. It provides a recommendation method that maximizes the satisfaction of all group members in consideration of the relationship with the group members, the professionalism of the members, and the tendency of the members. "Recommendation System Using Personalized Expert Group", Journal of Korean Institute of Intelligent Systems, Vol. 23, No. 1, pp7-11, 2013, , Which is used for the recommendation system [4]. "The Impact of the Characteristics of SNS Digital Environment on Intention to Purchase Cultural Contents", The Korean Society of Contents, Vol. 12, pp336-345, 2012, "Information on online and digital environments including SNS We analyzed the influence of items, groups, and various contents on recommendation or purchase.

Among the existing recommendation techniques, there have been a lot of researches recently on group recommendation techniques using collaborative filtering. Collaborative filtering is a technique that identifies other users with similar tendencies based on users' preferences and recommends information that is not present to the user among the information that other users have. Collaborative filtering can be broadly divided into user - based collaborative filtering and item - based collaborative filtering, and many related models have been studied. In this paper, we propose a recommendation system in cooperation with a collaborative filtering model. In recent years, research has been carried out on recommendation techniques considering information from experts in order to provide reliable recommendation to users. The basic meaning of an expert is defined as a person who has a considerable knowledge and experience in the field of studying or engaging in an area. In other words, considering the expert's information means that you can get a higher quality recommendation.

Existing group recommendation techniques generate preferences based on static profiles and recommend corresponding items or related groups. However, since most users do not change the profile entered at the time of joining the group, it is not possible to use the changed content of the user's recommendation. As a result, the accuracy of the recommendation is degraded and the latest information is not reflected. Also, since only the user profile information is reflected in the recommendation process, there is a problem that it can not reflect the recent tendency of other users.

[1] L. Quijano-Sanchez, J. A. Recio-Garc┱a, and B. Diaz-Agudo, "Personality and Social Trust in Group Recommendations", Proc. International Conference on Tools with Artificial Intelligence, Vol.3, pp.121-126, 2010. [2] L. Boratto, S. Carta, A. Chessa, M. Agelli, and M. L. Clemente, "Group Recommendation with Automatic Identification of Users Communities", Proc. International Workshop on Intelligent Web Interaction, Vol.3, pp.547-550, 2009. [3] M. Gartrell, X. Xing, Q. Lv, A. Beach, R. Han, S. Mishra, and K. Seada, "Enhancing Group Recommendation by Incorporating Social Relationship Interactions", Proc. International ACM SIGGROUP Conference on Supporting Group Work, pp. 97-106, 2010. [4] Jung Yeon Oh, Lee Sung Woo, Lee Ji Hyung, "A Recommendation System Using Personalized Expert Group", The Korea Institute of Intelligent Systems, Vol. 23, No. 1, pp. 7-11, 2013. [5] Lee, Han Suk, "The Influence of Characteristics of SNS Digital Environment on Purchase Intention of Cultural Contents", The Korea Contents Association, Vol. 12, No. 7 pp336-345, 2012.

SUMMARY OF THE INVENTION Accordingly, the present invention has been made to solve the above problems occurring in the prior art, and it is an object of the present invention to provide a method and apparatus for collecting user's recent group activity information and updating profile information, A group recommendation system and method in a social network environment that recommends a group using profile data of a group.

According to another aspect of the present invention, there is provided a group recommendation system for collecting group activity information of a user or collecting tendency information based on static profile information of a user to generate and update activity profile information A profile management unit; A group recommendation unit for recommending a group by comparing the preference and the group similarity by performing collaborative filtering through the similar user search and the expert search from the activity profile information, ; And a database for storing the activity profile information and the static profile information.

Wherein the profile management unit comprises: a group activity information collection unit for collecting the group activity information when a user is subscribed to the group; A tendency information collecting unit for collecting tendency information based on the static profile information when the user is not subscribed to the group; And an activity profile generating unit for generating and updating the activity profile information, which is dynamic profile information, based on an activity recently performed by the user.

The group recommendation unit may include a similar user searching unit for searching for a similar user having a similar tendency to a user; An expert search unit for determining an expert among the similar users; A collaborative filtering performing unit for performing collaborative filtering based on the activity profile information of the user and the activity profile information of the user derived by the expert; A preference keyword selection unit for generating a preference keyword from the information derived through the collaborative filtering; A similarity determination unit for comparing the generated preference keyword with similarity of the tag in the group and the keyword; And a list output unit for recommending a group having the highest degree of similarity.

Meanwhile, a group recommendation method in a social network environment of the present invention includes: (a) generating and updating activity profile information of a user; And (b) performing collaborative filtering through search of a similar user from the activity profile information to generate a preference, and recommending the group by comparing the preference and the group similarity.

In the step (a), the activity profile information may be generated by collecting and generating group activity information, or may be generated by collecting trend information through the user's static profile information. In the case of using the static profile information in the step (a) , It is preferable to search for and recommend a group through the static profile information.

The step (a) includes: classifying the activity groups by category based on the activity group information of each user; Calculating an amount of activity in which the activity in the group classified into the same category is quantified; And comparing the classified category with the previous profile information to finally generate and update the profile.

The step (b) comprises the steps of: selecting a user having a similar tendency to a user who operates in the same group category as the user; Performing collaborative filtering based on the user having the similar tendency; Extracting a keyword that is not present in the activity profile keyword item of the activity profile keyword item of the user having the similar tendency; Assigning a score corresponding to the degree of similarity with the user to the keyword; Selecting a preferred keyword when the given score is equal to or greater than a threshold value; And recommending a group corresponding to the preference keyword.

The step of performing the collaborative filtering further includes the step of discriminating experts among the users having the similar tendencies, wherein the discrimination of the experts is performed by users having an activity amount exceeding a threshold value in a group of users derived by similar users .

As described above, according to the group recommendation system and method in the social network environment according to the present invention, since the latest group activity information of the user is collected and the profile information is updated, the latest And it can provide diverse and accurate group recommendation service to the users than the existing technique because it considers the profile information of other users having similar tendencies in the field.

In addition, through collaborative filtering, users can be provided with more diverse groups by recommending groups using profile data of users (experts) similar to their own tendencies.

FIG. 1 is a diagram showing the overall configuration of a group recommendation system according to an embodiment of the present invention.
2 is a diagram illustrating an activity profile creation and update structure according to an embodiment of the present invention.
3 is a diagram illustrating an activity profile update algorithm according to an exemplary embodiment of the present invention.
4 is a diagram illustrating an activity amount calculation algorithm according to an embodiment of the present invention.
FIG. 5 is a flowchart illustrating an activity calculation process according to an embodiment of the present invention.
6 is a diagram illustrating an algorithm for generating an activity profile according to an exemplary embodiment of the present invention.
7 is a flowchart illustrating an activity profile updating process according to an embodiment of the present invention.
FIG. 8 is a diagram illustrating a group recommendation procedure according to an embodiment of the present invention.
9 is a diagram illustrating a user similarity extraction process according to an embodiment of the present invention.
FIG. 10 is a diagram illustrating an expert determination process according to an embodiment of the present invention.
11 is a diagram illustrating the use of expert information according to an embodiment of the present invention.
12 is a diagram illustrating a process of extracting a preferred keyword according to an embodiment of the present invention.
13 is a flowchart illustrating a recommendation priority extraction process according to an embodiment of the present invention.
14 is a diagram illustrating a group recommendation algorithm according to an embodiment of the present invention.

As the use of SNS (Social Network Service) increases rapidly, researches on recommendation techniques are actively being carried out. Recommendation techniques provide various services that users like or need in real time, It is a technique to provide appropriate groups based on propensity information.

In the present invention, as a group recommendation technique, a technique of recommending a group to a user using profile and collaborative filtering is presented. In other words, we propose a static group recommendation method based on the user dynamic profile and the tendency of users of social network. In the present invention, a static group is defined as an online community in which users gather for a specific purpose.

The group recommendation scheme proposed in the present invention improves the reliability and satisfaction of the group recommendation considering both the recent tendency of the user and the tendency of users in the group. It dynamically creates a profile through the user's activity history and reflects the recent user's tendency to change through continuous updating. In the group recommendation process, the activity information of the existing users in the group is collected, the inclination information of recent users is reflected in the group, and the profile of other users in the group is taken into consideration, thereby improving the reliability of the recommendation group. In the recommendation process, we recommend using groups of collaborative filtering to simultaneously consider other users and their inclination. That is, by applying collaborative filtering to the group recommendation technique, the satisfaction of the user is improved because another user who is similar to his / her own tendency is recommended in the group. In order to consider the expertise information of the group recommendation, if the partner user is an expert of the content, the profile information of the person is applied considering the weighted value. That is, by considering the information of experts in the group, a higher quality group than the generally recommended group is recommended.

Hereinafter, a group recommendation system and method in a social network environment according to the present invention will be described in detail with reference to the accompanying drawings.

FIG. 1 is a diagram showing the overall configuration of a group recommendation system according to an embodiment of the present invention.

Referring to FIG. 1, the group recommendation system of the present invention includes a profile management unit 1 for collecting group activity information of a user, collecting tendency information through static profile information of a user and generating and updating activity profile information, A group recommendation unit 2 for recommending a group by comparing the preference and the group similarity by performing collaborative filtering through similar user searching and expert search from the activity profile information, And a database 3 for storing activity profile information and static profile information.

The profile management unit 1 includes a group activity information collection unit 11 for collecting group activity information of a user when the user is registered in the group, A tendency information collecting unit 12 for collecting tendency information, and an activity profile generating unit 13 for generating and updating activity profile information, which is dynamic profile information, based on an activity recently performed by the user.

The group recommendation unit 2 includes a similar user search unit 21 for searching for similar users having similar tendencies to the user, a specialist search unit 22 for determining the presence or absence of an expert in the group, A collaborative filtering performing unit 23 for performing collaborative filtering based on a user's activity profile derived as a user, a preference keyword selecting unit 24 for generating a preference keyword from the information derived through collaborative filtering, A similarity determination unit 25 that compares the keyword with similarity of the tag and the keyword in the group, and a list output unit 26 that recommends the group having the highest similarity.

Here, it further includes a group search unit 27 using a static profile keyword for searching a group using a static profile keyword.

The group recommendation system of the present invention configured as described above determines whether or not the group is subscribed to the group for group recommendation. If the user is not in the group, he recommends the group using the static profile information described by the user. If the user is subscribed to the group, the user's preference is generated based on the activity recently performed by the user in the group, and the activity profile, which is the dynamic profile defined in the present invention, is generated. In order to derive recommendation information by performing collaborative filtering, a similar user having a tendency similar to the user is first searched. Search for similar users searches for similar users by comparing the user's activity profile with the activity profile information of the members in the group. Afterwards, we will judge the presence of experts to recommend a higher quality group in recommending a group. If there is an expert, the activity profile information of the expert performs collaborative filtering with a high weight. Collaborative filtering is based on the user's activity profile information and the user's activity profile derived as a similar user. After that, collaborative filtering is performed to generate the preference through the derived information. The preference is derived from the keyword that reflects the characteristics of the user and is defined as the preferred keyword. Through the generated preference keyword, group similarity is compared with the tag in the group and the group recommendation is performed.

A static profile is a direct description of what the user is interested in, and most users do not change that profile. However, since users tend to change their tendency with time, there is a problem in using the static profile for the purpose of grasping the user's recent tendency. Accordingly, in the technique proposed in the present invention, an activity profile is used to express a recent tendency of a user. The activity profile is the same as the dynamic profile that is periodically updated as a result of extracting the items of interest of the user based on the user's group activity information. When a user joins a group and starts an activity, the user enters a static profile, but the more active the group activity, the more updated the activity profile reflects the user's up-to-date behavior.

The activity profile update collects the group activity information in the same category and generates the user's propensity information for each user. For example, suppose that there are categories of 'music' and 'exercise' and that the user is active in three 'music' related groups and two 'exercise' related groups. The activity profile updates the recent tendency by synthesizing the activities performed by the three group information related to the music in the music related tendency information, and the activity related information includes the activity information To update the activity profile.

2 is a diagram illustrating an activity profile creation and update structure according to an embodiment of the present invention.

Referring to FIG. 2, a group in which a user is active is classified into four categories of music, movement, game, and language. Groups 1 and 2 are classified into music categories, while groups 3 and 4 are classified into exercise categories. Groups 5 and 6 are categorized into a game category, and groups 7 and 8 are classified into a language category. Therefore, the update of the activity profile finally generates and updates the activity profile by synthesizing the information of group 1 and group 2, group 3 and group 4, group 5 and group 6, group 7 and group 8 included in the same category category.

3 is a diagram illustrating an activity profile update algorithm according to an exemplary embodiment of the present invention.

Referring to FIG. 3, the activity group information of each user is input to generate and update the profile for the first time (S1). Then, the activity group is classified into categories based on the inputted activity group information (S2). The classified categories are integrated (S3) and compared with the previous profile information (S4). Finally, a profile is generated and updated (S5).

내에서 사용자의 활동량

(Profile Value)은 그룹에서 사용자

가 수행한 활동량을 나타낸다. (식 1)에서

는 그룹

내에서

사용자가 정의한 임계 시간 내 수행한 활동량을 계산하기 위한 속성,

는

에 활동에 영향을 줄 수 있는 속성 값에 부여된 가중치,

는 현재 시각,

은 사용자의 활동량 정보가 수집되는 최초시점,

은 수집되는 최초시점을 조절하는 변수이다.

와

는 과거부터 현재까지 활동량을 종합할 때 각각의 시점에서 부여될

값의 가중치를 조정하기 위한 상수이다. 예를 들어,

와

에 의해서 수식의 기울기 증가율이 높아지면 현재 시점에 가까운 과거 활동정보의 가중치는 높아지며 기울기의 증가율이 낮다면 현재 시점에서 가까운 과거 활동정보의 가중치는 낮아진다.
To update the activity profile, first calculate the amount of activity for each group to which the user subscribes. The amount of activity is a numerical value of the activity of the user in the group that has recently been classified in the same category. The amount of activity is calculated by collecting the attribute value that can affect the activity performed by the user in the group for a certain period of time. At this time, the attribute value may correspond to the written article, the comment, the number of check-in, and the like. The activity of each user is calculated using (Equation 1). group

The amount of user activity within

(Profile Value)

Of the activity. (Equation 1)

Group

Within

An attribute for calculating the amount of activity performed within a user-defined critical time,

The

A weight given to an attribute value that can affect the activity,

The current time,

The first time point at which the user's activity amount information is collected,

Is a variable that adjusts the earliest point of time that is collected.

Wow

Will be given at each time point when synthesizing the amount of activity from past to present

This is a constant for adjusting the weights of the values. E.g,

Wow

, The weight of the past activity information near to the current point is higher and the weight of the past activity information near the current point is lowered if the slope increase rate is low.

--- (식 1)

--- (1)

4 is a diagram illustrating an activity amount calculation algorithm according to an embodiment of the present invention.

를 설정한다(S14). 속성 값이 결정되면 속성 값 마다 적용시킬 가중치

를 결정하고(S15) 최종적으로 사용자의 그룹 활동량

를 계산한다(S16). Referring to FIG. 4, the activity amount is calculated based on the group information and the activity history information of the user in order to calculate the activity amount. In order to calculate the activity amount of the first user for each category, the group information on which the user is active is classified by category (S11). Then, a period for collecting the activity amount of the user is set (S12). When the collection period is set, a weight to be applied is set because the weight of the amount of activity applied to each period is different (S13). When the weight is set, the attribute value

(S14). If the attribute value is determined, the weight to be applied to each property value

(S15). Finally, the user's group activity amount

(S16).

FIG. 5 is a flowchart illustrating an activity calculation process according to an embodiment of the present invention.

Referring to FIG. 5, an example of a process of calculating a user activity amount is assumed. It is assumed that a user performs an activity on an attribute value corresponding to (a) in a specific group. It is also assumed that the amount of activity calculation is based on the amount of activity within the last four months. It is assumed that the weight given to each property for calculating the activity amount is the same as (b), and the weight for each viewpoint is the same as (c). In the case of January, ((50 * 0.5) + (90 * 0.3) + (200 * 0.2)) * 0.1 is finally obtained. The calculations for February, March and April are the same, and the amount of activity in this group for the last four months is 53.

값의 총합에서 높은 값을 가지는

개의 그룹의 키워드 정보를 사용자의 활동 프로필을 생성하기 위해 사용한다. 그룹 키워드는 그룹의 특성을 설명하기 위해 부여한 값으로 그룹 관리자에 의해 부여되거나 텍스트 마이닝을 통해 자동으로 생성된다. 활동 프로필은 사용자가 활동한 그룹 카테고리별로 생성된다. 동일한 그룹 카테고리에 포함된 각 그룹 키워드들에서 활동 프로필로 사용하기 위한 키워드를 선출하기 위해 각 그룹 키워드에

를 부여한다.

는 그룹

에 포함된 키워드

에 부여된 사용자

의

으로 각 키워드에 부여된 가중치를 의미한다. 동일한 그룹 카테고리에 포함된 모든 그룹에서

가 부여되면 (식 2)를 이용하여 사용자

에 대한 키워드

의 선호도를 생성한다. 키워드에 대한 선호도가 생성되면 임계치 이상의 선호도를 갖는 키워드를 활동 프로필로 생성한다. 즉, 사용자가 많은 활동한 그룹 내에 포함된 키워드들을 사용자의 활동 프로필로 사용한다.
The activity profile is created based on the user's activity information when the group in which the user is active exists. Combine all activity performed in the same group category to create a user's activity profile. Calculated

Higher values in the sum of values

Keyword information of the group is used to generate the user's activity profile. The group keyword is a value assigned by the group manager to describe the characteristics of the group or is automatically generated through text mining. An activity profile is created for each group category in which the user is active. To select keywords for use as activity profiles in each group keyword in the same group category,

.

Group

Keywords included in

User granted to

of

And a weight value given to each keyword. In all groups in the same group category

(Equation 2), the user

Keywords for

The preference of < / RTI > When a preference for a keyword is generated, a keyword having a preference higher than a threshold value is generated as an activity profile. In other words, the user uses the keywords included in the group in which the user has many activities as the activity profile of the user.

--- (식 2)

--- (Equation 2)

6 is a diagram illustrating an algorithm for generating an activity profile according to an exemplary embodiment of the present invention.

값을 생성한다(S23).

값이 도출되면 내림차순으로 정렬을 실시하며(S24) 최종적으로 상위 n개의 키워드를 활동 프로필의 키워드로 사용하게 된다(S25 ~ S26).Referring to FIG. 6, an activity group is loaded to create a profile creation for the first time. When the activity group is loaded, each group is classified by category (S21). The classified activity group derives the keywords of the classified group to derive the keyword to be used in the activity profile (S22). Then, the user activity amount PV value of the corresponding group is assigned to each keyword as a weight value

(S23).

When the values are derived, sorting is performed in descending order (S24). Finally, the top n keywords are used as keywords of the activity profile (S25 to S26).

As the user's tendency changes, the group activity also changes based on the tendency. Therefore, the activity profile is updated based on the group activity information.

7 is a flowchart illustrating an activity profile updating process according to an embodiment of the present invention.

가 운동과 어학에 관련된 2개의 카테고리에 활동하고 사용자의 활동량은 (식 1)을 이용하여 (a)와 같이 생성되었다고 가정하기로 한다. 각각의 그룹에 대한 활동량이 25, 45, 50, 60으로 부여 되었기 때문에 그룹 1 키워드 가중치는 가중치 25, 그룹2 키워드 가중치는 45, 그룹3 키워드 가중치는 50, 그룹4 키워드의 가중치는 60으로 부여된다. 결과적으로 키워드 가중치는 축구70, 배구 70, 농구 45, 족구 25, 테니스 25로 부여된다. 키워드 임계치가 40이라 가정한다면 (b)와 같이 이전 활동 프로필의 운동 카테고리의 키워드로 부여된 축구, 농구의 키워드가 가장 많이 겹치면서 가장 높은 가중치를 가지는 축구, 배구, 농구 키워드로 갱신된다. 이와 유사하게 어학 카테고리의 활동 프로필도 중국어, 독일어, 일어로 변경된다.
Referring to Figure 7,

Is active in two categories related to exercise and language, and the amount of activity of the user is assumed to be generated as in (a) using (Equation 1). The group 1 keyword weight is assigned a weight of 25, the group 2 keyword weight is 45, the group 3 keyword weight is 50, and the weight of a group 4 keyword is 60, because the activity amounts for each group are given as 25, 45, 50, . As a result, the keyword weight is assigned to the soccer 70, the volleyball 70, the basketball 45, the football 25, and the tennis 25. Assuming that the threshold of the keyword is 40, as shown in (b), the keywords of soccer and basketball, which are keywords of the exercise category of the previous activity profile, are renewed with the highest weighted soccer, volleyball and basketball keywords. Similarly, the activity profile of the language category is changed to Chinese, German, and Japanese.

In order to reflect the tendency of other users working in the same group category as the user, the proposed technique performs collaborative filtering on the activity profile. In other words, a preference keyword is generated in consideration of the activity profile indicating the user's tendency and the tendency of users having similar tendencies existing in the group for group recommendation. In order to perform collaborative filtering, a user having a similar tendency is first selected for a user who operates in the same group category as the user. If a user with similar inclinations is derived, he or she identifies an expert among similar users. Expert discrimination is selected among users who are derived from similar users and who have activity levels above the threshold. Collaborative filtering is then performed based on similar users derived later. Collaborative filtering is performed to extract keywords that do not exist in the activity profile keyword items of the similar user's activity profile keyword items. In order to prioritize the extracted keywords, a score is assigned to each keyword according to the degree of similarity with the user. If the score is more than the threshold value, it is selected as a preferred keyword, and the group is recommended.

FIG. 8 is a diagram illustrating a group recommendation procedure according to an embodiment of the present invention.

Referring to FIG. 8, in a first step, the user's recent activity group information is collected and analyzed from the database 3. In the second step, similar users are derived by comparing and analyzing the number of profile keywords based on the collected information. In step 3, user selection is performed to distinguish between expert information and general information when collaborative filtering is performed. Among the derived similar users, users having an activity level equal to or greater than the threshold value are selected as experts and the remaining users are classified as general similar users. In step 4, collaborative filtering is performed based on the information of the classified users. By performing collaborative filtering at the same time as the profiles of users who are selected as experts, profiles of general similar users, and profiles of users to be recommended, the most appropriate keywords are extracted. In the fifth step, the group is searched through extracted keywords, and finally, the most appropriate group is recommended to the user.

는 그룹 카테고리

에서 사용자

,

의 활동 프로필 유사성,

는 그룹 카테고리 k에서 사용자

의 활동 프로필 키워드 수,

는 그룹 카테고리 k에서 사용자

,

에 대한 공동 활동 프로필 키워드 수이다.
To select users with a similar tendency to the user, search for users who are active within the same group category as the user and compare the similarity of the users' activity profiles. Similarity is calculated by comparing the keywords contained in the activity profile of the user who has been active in the same group category. (Equation 3) is an equation for calculating the similarity of users. At this time,

Group category

From

,

Activity profile similarity,

Users in group category k

Activity profile Keywords,

Users in group category k

,

The number of collaborative activity profile keywords for.

--- (식 3)

--- (Equation 3)

9 is a diagram illustrating a user similarity extraction process according to an embodiment of the present invention.

Referring to FIG. 9, similarity between the user B, the user C, the user D, and the user E is calculated based on the activity profile of the user A (Equation 3). For example, as shown in (a), user B includes three identical keywords with user A, and users C, D, and E include the same keywords 2, 1, and 2, respectively. When user similarity is calculated, the similarity of user B is about 0.56 and the users C, D, and E are about 0.33 to 0.06 as shown in (b). Assuming that the similarity threshold is 0.5, the user determined to be a similar user is user B.

The existing group recommendation method recommends the group to the user based on the individual propensity information. However, existing techniques do not guarantee quality or reliability of the recommended group. In the present invention, the information of experts is considered to provide users with a higher quality and more reliable group when recommending groups. The basic meaning of an expert is defined as a person who has a considerable knowledge and experience in the field of research or work in an area. Therefore, in the present invention, it is assumed that the expert has higher quality and more reliable data than general users.

The experts in this category select the most active n people in the category and select them as experts in the group. Also, the selection of experts is selected within the range of users derived from users similar to themselves. The activity amount of the expert is extracted by using PV, and the extracted PV values are sorted and used to perform collaborative filtering by selecting users having a value equal to or higher than a threshold selected by the user as an expert.

FIG. 10 is a diagram illustrating an expert determination process according to an embodiment of the present invention.

Referring to FIG. 10, it is assumed that three users who are determined to have similar tendencies in the same category group are derived. To derive the expert from the derived user, calculate the activity quantity PV of the activity group within the same category category previously. Assuming that the PV values of each user are 50, 70, and 90, and the user sets a PV value of 80 as a specialist, the user 3 who satisfies the condition is selected as the expert of the group.

Users who are selected as experts will perform the collaborative filtering process using the respective activity profile information. Apply weight to information in the activity profile of the user selected as an expert before performing the collaborative filtering process. Since experts have more information than ordinary users, they perform collaborative filtering by applying higher weight than general users.

11 is a diagram illustrating the use of expert information according to an embodiment of the present invention.

Referring to FIG. 11, it is assumed that a user who has been selected as a general user and an expert is given an example of a weighting process, and an expert user having a similar tendency and similarity to a general user having a similar tendency is discriminated. In this case, when the collaborative filtering is performed, the weight of the information is 100%, 60% for the expert user, and 40% for the general user.

는 사용자

에 대해 그룹 카테고리

에서 협업 필터링을 통해 추출한 키워드

의 점수를 나타내며

은 유사 사용자 수를 나타낸다. 추출된 키워드에 우선 순위를 부여하기 위하여 (식 4)를 통해 도출된 유사성 값을 이용하여 최종 키워드 점수를 부여한다. 점수가 부여된 키워드 중 임계치 이상의 값을 가지는 키워드를 선호 키워드로 선정한다.
Collaborative filtering is used to recommend groups by reflecting the tendencies of other users with similar tendencies to specific users. That is, user activity profile having similar tendency is extracted through collaboration filtering and group recommendation is utilized. Use the preferred keyword to search for the target group for group recommendation. The preference keywords are extracted from the activity profiles of users having a similar tendency within the same group category, such as those in which a particular user is likely to be interested. Priority is assigned to the extracted keywords to define them as preferred keywords. The user's activity profile extracted through (Equation 4) and the activity profile of the user to be recommended for the group are collocated and filtered to extract the preferred keywords. At this time,

User

About group categories

Keywords extracted from collaborative filtering in

Is the score of

Represents the number of similar users. In order to assign the priority to the extracted keywords, the final keyword score is given using the similarity value derived from (Equation 4). A keyword having a value equal to or larger than a threshold value among the keywords to which a score is assigned is selected as a preferred keyword.

--- (식 4)

--- (Equation 4)

12 is a diagram illustrating a process of extracting a preferred keyword according to an embodiment of the present invention.

을 계산하는 과정을 나타낸 것으로서, 운동에 관련된 그룹에서 협업 필터링을 통해 추출된 키워드가 '테니스', '탁구', '하키'라고 가정하기로 한다. 테니스의 경우 사용자 B의 유사도 점수가 0.56, 사용자 C의 경우 0.33, 사용자 E의 경우 0.33이기 때문에 최종적으로 테니스 키워드의 KP값은 1.22이다. 임계치 값이 1이라고 가정할 경우 최종적으로 '테니스' 키워드가 선호 키워드로 선정된다.12,

, And it is assumed that the keywords extracted through collaborative filtering in a group related to exercise are 'tennis', 'table tennis', and 'hockey'. In the case of tennis, since the score of similarity of user B is 0.56, that of user C is 0.33, and that of user E is 0.33, the KP value of tennis keyword is finally 1.22. If the threshold value is 1, the keyword 'tennis' is finally selected as a preferred keyword.

When a preferred keyword is selected, a recommendation group is searched for by a user. The recommendation ranking is determined according to the number of matching keywords by comparing the keywords assigned to the group with the preferred keywords.

13 is a flowchart illustrating a recommendation priority extraction process according to an embodiment of the present invention.

Referring to FIG. 13, it is assumed that a group is recommended based on a recommendation keyword, and it is assumed that the preferred keywords selected through Formula 4 are 'tennis', 'table tennis', and 'hockey'. It is assumed that groups A through D are derived as a result of searching for a group through a preferred keyword. The priority of the recommended group is determined based on the number of matching of the preferred keyword and the group keyword. As a result, the group recommendation priority is determined in the order of groups B, C, A,

14 is a diagram illustrating a group recommendation algorithm according to an embodiment of the present invention.

Referring to FIG. 14, the input is a Profile_SET, which is information about a profile, and an output is a Group_SET, which is information about a recommended group. The user who starts the group activity for the first time performs the group recommendation using the information of the static profile (S32 to S34). In the case of a user having an activity group, the activity profile is extracted through the group activity history, and the user is searched for the group to which the user belongs, and the group is classified according to the category (S36 to S38). When a group is classified, a user having a similar tendency in the group is searched through (Equation 3). Then, similar nouns are selected from the similar users as experts, and the keywords existing in the users having similar tendencies are derived and the scores are given (S39 to S46). A preferred keyword having a score of a threshold value or more among the keywords to which a score is assigned is extracted and compared with the group information to recommend a group matching the largest number of keywords (S48 to S52).

While the present invention has been particularly shown and described with reference to exemplary embodiments thereof, it is to be understood that the invention is not limited to the disclosed exemplary embodiments, but, on the contrary, is intended to cover various modifications and equivalent arrangements included within the spirit and scope of the invention.

1: Profile manager
2: Group recommendation section
3: Database

Claims (14)

delete delete delete (a) a profile management unit generating and updating user activity profile information; And
(b) performing a collaborative filtering by searching for a similar user from the activity profile information to generate a preference, and recommending the group by comparing the preference with the group similarity,

The step (a)
Categorizing the activity groups into categories based on the activity group information of each user;
Calculating an amount of activity in which the activity in the group classified into the same category is quantified; And
Collecting the classified categories and comparing them with the previous profile information to finally generate and update the profile,

The step (b)
Selecting a user having a similar tendency to a user who operates in the same group category as the user;
Performing collaborative filtering based on the user having the similar tendency;
Extracting a keyword that is not present in the activity profile keyword item of the activity profile keyword item of the user having the similar tendency;
Assigning a score corresponding to the degree of similarity with the user to the keyword;
Selecting a preferred keyword when the given score is equal to or greater than a threshold value; And
And recommending a group corresponding to the preference keyword,

Wherein the collaborative filtering is performed by a user who has similar tendencies,
The expert may be a group of users derived from similar users

Users within

The amount of activity

) Is calculated using Equation (1) below, and the users are selected as the users whose activity amount is equal to or higher than the threshold value,
A group recommendation method in a social network environment in which a weight different from the general user is applied to the activity profile information of the user selected as the expert.

--- (1)
here,

Group

Within

An attribute for calculating the amount of activity performed within a user-defined critical time,

The

A weight given to an attribute value that can affect the activity,

The current time,

The first time point at which the user's activity amount information is collected,

Represents a variable that adjusts the initial point of time of collection.

Wow

Will be given at each time point when synthesizing the amount of activity from past to present

This is a constant for adjusting the weights of the values.
5. The method of claim 4,
The group recommendation method in a social network environment in which the activity profile information in the step (a) is collected by generating group activity information, or is generated by collecting trend information through static profile information of a user.
6. The method of claim 5,
Wherein the group recommendation method searches for and recommends a group through the static profile information when the static profile information is used in the step (a).
delete delete 5. The method of claim 4,
user

Keywords for

The above preference (

) Is a group recommendation method in a social network environment generated using (Equation 2) below.

--- (Equation 2)
here,

Group

Keywords included in

User granted to

of

And a weight value given to each keyword.
delete 5. The method of claim 4,
The user's group category

From

,

Activity profile similarity (

) Is a group recommendation method in a social network environment which is calculated by (Equation 3) below.

--- (Equation 3)
here,

Users in group category k

Activity profile Keywords,

Users in group category k

,

And the number of cooperative activity profile keywords.
delete delete 5. The method of claim 4,
user

About group categories

Keywords extracted from collaborative filtering in

Score of

) Is a method for group recommendation in a social network environment in which a final keyword score is given using the similarity value derived from (Equation 4) below, and a keyword having a given score is selected as a preferred keyword.

--- (Equation 4)
here,

Represents the number of similar users,

Group category < RTI ID = 0.0 >

From

,

Of activity profile similarity.

Images