KR20120106667A - Method and apparatus for producing user reputation of online network - Google Patents

Abstract

PURPOSE: A method for calculating a reputation of an online network user and a device thereof are provided to calculate reputations of online network users by considering the number of user-centric feedbacks regardless of the number of postings written by each user. CONSTITUTION: A receiving unit(800) receives feedback information about online network postings. A first information obtaining unit(810) obtains first information for providing a feedback to a first user. An initial reputation calculating unit(820) calculates an initial reputation score of the first user by using the first information. A second information obtaining unit(830) obtains second information for exchanging feedbacks between the first user and other users. A sociality calculating unit(840) calculates a sociality score of the first user by using the second information. [Reference numerals] (800) Receiving unit; (810) First information obtaining unit; (820) Initial reputation calculating unit; (830) Second information obtaining unit; (840) Sociality calculating unit; (850) Pure reputation calculating unit; (860) Reducing unit; (870) Final reputation calculating unit

Description

Method and device for calculating reputation of online network user {Method and Apparatus for Producing User Reputation of Online Network}

The present invention relates to a method and apparatus for calculating reputation of an online network user. Specifically, the present invention relates to a method and apparatus for receiving feedback information on an online network post written by an online network user and calculating a reputation of an online network user in consideration of sociality based on feedback between online network users.

Due to the development of the Internet, various types of online networks such as an online community, a blogosphere, and a social network are actively operated. Internet users can create various types of posts such as user-generated contents (UGC) to user-created contents (UCC) and publish them to other users.

The technique of sorting these various posts by ranking according to a predetermined criterion and searching for a desired post among them has also developed along with these posts or content technologies. The document ranking technique is used to search for such posts based on the ranking. The conventional document ranking technique can be broadly divided into a keyword ranking algorithm based on a keyword-based document ranking algorithm and a link analysis between web documents.

One of the keyword-based document ranking algorithms is based on the traditional theory of information retrieval, TF-IDF (Term Frequency-Inverse Document Frequency), which is the number of words in a document and the number of documents that contain words. An inverse document frequency of document frequency is used.

On the other hand, among the document ranking algorithms by link analysis between web documents, the Page Rank algorithm analyzes the cross-link relationship between web documents based on a random walk model, thereby ranking the global rank of each web document. It is an algorithm that calculates. The Hyperlink-Induced Topic Search (HITS) algorithm analyzes the link relations of the web document search results and calculates two ranks for the authority and the hub of each document. Briefly, the conventional researches related to the page rank algorithm and the HITS algorithm are as follows.

P.Jurczyk and E.Agichtein.Discovering authorities in question answer communities by using link analysis.CIKM 2007 finds authority and herb in Yahoo! 'S Answers The HITS algorithm was used to identify the authoritative users who answered a lot of good questions and the hub users who asked a lot of good questions. In Z.Gyogyi and G.Koutrika's papers (Z.Gyogyiand G.Koutrika, J.Pedersen and HG-Molina. Questioning Yahoo! Answers.WWW2008Workshop-QAWeb2008, 2008) The techniques used are discussed.

Meanwhile, Ko Fujimura et al. (Ko Fujimura, Takafumi Inoue and Masayuki Sugisaki.The EigenRumor Algorithm for Ranking Blogs. A. Kristikopoulos et al. (A. Kristikopoulos, M. Sideri, and I. Varlamis. Blgrank: ranking weblogs based on connectivity and similarity features.Proc. Of AAA-IDEA, 2006) use implicit hyperlinks. We proposed a blog ranking technique. For example, assume that two blog posts with the same topic are implicitly linked by hyperlinks. These implicit hyperlinks can compensate for sparse link relationships between blog posts. Implicit hyperlinks apply to bloggers and commenters on the same topic in addition to the same topic. In addition, J.Zhang et al. (J.Zhang, MS Ackerman and L. Adamic. Expertise Networks in Online Communities: Structure and Algorithms.WWW 2007) also suggest a pagerank and HITS algorithm to estimate user expertise in online networks. Was used.

The limitation that the existing page ranking algorithm and the ranking algorithm using the HITS algorithm is a document ranking algorithm by link analysis between web documents, which is mainly used to calculate the ranking of posts and may be applied to calculate the user's ranking. There was.

Meanwhile, users of the online network can leave feedback on published posts. The user's feedback on the post may take the form of comments, comments, empathy, recommendations, trackbacks, scraps and links. In other words, in an online network, users write posts, give feedback to and interact with posts from other users. In an online network, a user may provide feedback to posts of other users with similar interests.

In addition, by exchanging interactions between users on the online network, or by intimate relationships on-line, the users on the online network can form social relationships among users, such as registering some of the other users as friends. In this case, the second user who is registered as a friend of the first user is notified by the online network when the first user creates or updates a post, and visits the first user's space (e.g., blog or homepage). It becomes possible.

The second user who visits the space of the first user on the online network includes a user who has already formed a social relationship with the first user and a user who has never formed a social relationship with the first user. The second user who has already formed a social relationship with the first user does not consider the benefits, professionalism, etc. of the content of the first user's post, but simply forms a social relationship with the first user. You can also visit. That is, when the sociality of the first user is high (for example, when there are many registered friends of the first user), the first user is independent of the post of the first user, that is, despite the poor contents of the post, There may be a problem that can receive a high reputation.

In this regard, the page rank algorithm and the HITS algorithm, which are the conventional document ranking techniques described above, have a limitation in overlooking the fact that the sociality of the user affects the reputation of each user. Therefore, there is a problem in that a fair user ranking cannot be calculated when a ranking is calculated between users using a page rank algorithm and an HITS algorithm without considering the nature of feedback sent and received on the online network.

In order to solve the above problems, an aspect of the present invention is to provide a method and apparatus for calculating the reputation of an online network user in consideration of the number of feedback provided by the user, regardless of the number of posts written by each user. do.

In order to solve the above problems, another aspect of the present invention is to provide a method and apparatus for receiving feedback information for an online network post and calculating reputation of an online network user in consideration of sociality based on feedback between online network users. The purpose.

In addition, an object of the present invention is to separate the reputation and sociality of online network users to define each and provide an effective calculation method.

In addition, an object of the present invention is to provide a fair and efficient ranking calculation method by separating reputation and sociality of online network users.

In order to achieve the above object, a first aspect of the present invention is a reputation calculation method of an online network user implemented in a reputation calculation apparatus, wherein the reputation calculation apparatus is an online network post created by a plurality of online network users at a reception unit. Receiving the feedback information for the first step, the first information obtaining unit obtains the first information which is the probability that each user to provide feedback to the first user of the user, the reputation score calculator using the first information A method of calculating a reputation of an online network user performing the step of calculating a reputation score of a user is provided.

The acquiring of the first information may include acquiring the number of feedbacks provided by the second user to the first user by dividing the number of feedbacks provided by the second user to the entire user. Provides a method for calculating a user's reputation.

The calculating of the reputation score of the first user may include performing a multiplication of the reputation information score of the second user among the first information and the user.

In addition, the calculating of the reputation score of the first user provides a method for calculating the reputation of an online network user comprising the step of calculating using Equation 1.

[Equation 1]

는 제 2 사용자가 작성한 피드백의 전체 개수, d는 감쇄 인자(damping factor))Where u is the first user, v is the second user, RR _rw (u) is the reputation score of the first user, RR _rw (v) is the reputation score of the second user, and b (u) is the first user. The set of users who provided the feedback, f (v) is the set of users who received the feedback written by the second user, F _{v- > u} is the number of feedbacks provided by the second user to the first user,

Is the total number of feedbacks written by the second user, d is the damping factor

In order to achieve the above object, a first aspect of the present invention is a reputation calculation method of an online network user implemented in a reputation calculation apparatus, wherein the reputation calculation apparatus is an online network post created by a plurality of online network users at a reception unit. Receiving the feedback information for the step, obtaining the first information which is the probability that each user in the first information providing feedback to the first user of the user, the first reputation score calculator using the first information Calculating an initial reputation score of the first user, acquiring second information which is a probability that the first user and each user exchange feedback in the second information obtaining unit, and using the second information in the social score calculating unit Calculating a sociability score of the first user and a first reputation score calculation unit. A method of calculating a reputation of an online network user performing a step of calculating a pure reputation score of a first user using an initial reputation score and a sociality score of a user.

The feedback also provides a method for calculating the reputation of an online network user including one or more of comments, comments, empathy, recommendations, trackbacks, scraps and links.

The acquiring of the first information may include acquiring the number of feedbacks provided by the second user to the first user by dividing the number of feedbacks provided by the second user to the entire user. Provides a method for calculating a user's reputation.

The calculating of the first reputation score of the first user may include performing a multiplication of the first information and the first reputation score of the second user among the users.

In addition, the step of calculating the first reputation score of the first user provides a method for calculating the reputation of an online network user comprising the step of using the equation (1).

[Equation 1]

는 제 2 사용자가 작성한 피드백의 전체 개수, d는 감쇄 인자(damping factor))Where u is the first user, v is the second user, RR _rw (u) is the first reputation score of the first user, RR _rw (v) is the first reputation score of the second user, and b (u) is the first The set of users that provided feedback to the user, f (v) is the set of users that received feedback generated by the second user, F _{v- > u} is the number of feedbacks provided by the second user to the first user,

Is the total number of feedbacks written by the second user, d is the damping factor

The calculating of an initial reputation score of the first user may include calculating an authority score of the first user by defining a second user as a hub node and a first user as an authority node, and a first user. Comprising a step of calculating the first reputation score of the first user according to the authority score of the online network user provides a method for calculating the reputation.

The calculating of the authority score of the first user may include calculating a second user hub score according to the initial authority score of the first user, and calculating a product of the first information and the second user hub score. It provides a method for calculating the reputation of an online network user comprising the step of calculating the authority score of the user.

The acquiring of the second information may include obtaining the second information by dividing the number of feedbacks exchanged with the second user among the users by the number of feedbacks exchanged with the entire user by the second user. Provides a method for calculating a user's reputation.

The number of feedbacks exchanged with the second user by the first user may include a number of first feedbacks using the number of feedbacks provided by the first user to the second user and feedbacks provided by the second user to the first user. It provides a method for calculating the reputation of an online network user calculated by using the second feedback number using the number of.

The number of feedbacks exchanged with the second user by the first user may include arithmetic mean, geometric mean, and harmonic mean of the smaller one of the first feedback number and the second feedback number, the second feedback number, and the second feedback number. It provides a method for calculating the reputation of an online network user calculated by one of the methods.

The first feedback count and the second feedback count are provided in consideration of the number of posts written by the first user and the number of posts written by the second user, respectively.

In addition, the first feedback number is calculated by multiplying the number of feedbacks provided by the first user to the second user and the average number of feedbacks provided by the first user to the second user per post, and the second feedback number 2 provides a method for calculating an online network user's reputation calculated by multiplying the number of feedbacks provided by a user to a first user and an average number of feedbacks provided to a first user by a second user per post.

The first feedback number is calculated by multiplying a log value of the number of feedbacks provided by the first user to the second user and the average number of feedbacks provided by the first user to the second user per post, and the second feedback. The number provides a method for calculating an online network user's reputation calculated by multiplying a log value of the number of feedbacks provided to the first user by the second user and the average number of feedbacks provided to the first user by the second user per post.

The calculating of the social score of the first user may include providing a multiplicity of the second information and the social score of the second user among the users.

In addition, the calculating of the social score of the first user provides a method for calculating the reputation of an online network user comprising calculating using Equation 2.

&Quot; (2) "

는 제 2 사용자가 교환한 피드백의 전체 개수, d는 감쇄 인자(damping factor))(Where u is the first user, v is the second user, SR (u) is the social score of the first user, SR (v) is the social score of the second user, and bd (u) provides feedback with the first user) The set of users exchanged, bd (v) is the set of users who exchanged feedback with the second user, F ^(s) _{v-> u} is the number of feedbacks exchanged between the first user and the second user,

Is the total number of feedbacks exchanged by the second user, d is the damping factor

The calculating of the net reputation score of the first user may include subtracting the social score of the first user from the initial reputation score of the first user. .

The calculating of the pure reputation score of the first user may include: subtracting the number of feedbacks exchanged with the second user by the first user from the number of feedbacks provided by the first user to the second user among the users; A method of calculating a reputation of an online network user, comprising: calculating a large value among predetermined thresholds and defining the calculated value as a pure reputation strength from a first user to a second user.

In addition, the calculating of the pure reputation score of the first user provides a method for calculating the reputation of an online network user comprising the step of using the equation (3).

&Quot; (3) "

는 제 1 사용자로부터 제 2 사용자로의 순수 평판도 강도,

는 제 2 사용자의 순수 평판도 강도의 합, d는 감쇄 인자(damping factor))(Where u is the first user, v is the second user among users, PRrw (u) is the pure reputation score of the first user, PRrw (v) is the pure reputation score of the second user, and b (u) is the first user) The set of users that provided feedback to the user, f (v) is the set of users that provided feedback by the second user,

Is the pure reputation strength from the first user to the second user,

Is the sum of the pure reputation strengths of the second user, d is the damping factor

In addition, the reputation calculator calculates the final reputation score of the first user by using the net reputation score of the first user and the social score of the first user in the final reputation score calculation unit of the online network user Provide a method for calculating reputation.

In addition, calculating the final reputation score provides an online network user reputation calculation method comprising the step of calculating using the equation (4).

&Quot; (4) "

(Where R _comb (u) is the final reputation score of the user u, PR (u) is the pure reputation score of the user u, SR (u) is the sociality score of the user u, a is a constant between 01)

In order to achieve the above object, a third aspect of the present invention provides a receiver for receiving feedback information on an online network post written by a plurality of online network users, and a first probability in which each user provides feedback to a first user among the users. A reputation information calculating device of an online network user includes a first information obtaining unit obtaining information and a reputation score calculating unit calculating a reputation score of a first user using the first information.

In order to achieve the above object, a fourth aspect of the present invention provides a receiver for receiving feedback information on an online network post written by a plurality of online network users, and a first probability in which each user provides feedback to a first user among the users. A first information acquisition unit for acquiring information, an initial reputation score calculator for calculating an initial reputation score of the first user by using the first information, and second information which is a probability that each of the first user and the user exchanges feedback A second information obtaining unit configured to calculate a social score of the first user using the second information, a social score score unit, and a first reputation using the initial reputation score and the social score of the first user to calculate the pure reputation score of the first user An on-line network user's reputation calculation apparatus including a pure reputation score calculation unit is provided.

In addition, the feedback provides an apparatus for calculating a reputation of an online network user including at least one of a comment, a comment, a sympathy, a recommendation, a trackback, a scrap, and a link.

The pure reputation score calculator may further include a reputation calculator for an online network user including a subtractor configured to subtract the social score of the first user from the first reputation score of the first user.

The present invention also provides an apparatus for calculating a reputation of an online network user further comprising a final reputation score calculator configured to calculate a final reputation score of the first user by using the pure reputation score of the first user and the sociality score of the first user.

According to an aspect of the present invention, regardless of the number of posts written by each user, the reputation of an online network user may be calculated in consideration of the number of feedbacks provided by the user.

According to an aspect of the present invention, by considering the feedback of other users to a user in the online network, there is an effect that can provide a reputation of the user in consideration of sociality between users.

In addition, it is possible to provide a fair and efficient ranking calculation method by separation of reputation and sociality of online network users.

1 is a graph model depicting feedback activity for a user's post.
FIG. 2 is a graph model reconstructing the feedback activity of FIG. 1 in accordance with the present invention.
3 is a graphical model illustrating the amount of interaction between users in accordance with the present invention.
4 is a graph model classifying relationships between online network users into sociality and reputational relationships according to the present invention.
5 is a graph model showing the social strength between users in accordance with the present invention.
6 is a graph model showing the pure plate strength between users in accordance with the present invention.
7 is a flowchart illustrating a schematic flow of a reputation calculation method of an online network user implemented in a reputation calculation apparatus according to the present invention.
8 is a block diagram illustrating a schematic configuration of an apparatus for calculating a reputation of a network user according to the present invention.
9 is a graph showing a SLRCLUB user score distribution according to the present invention.
10 is a graph illustrating a relationship between ranks of SLRCLUB users according to the present invention.
11 is a graph illustrating a user having a high rank among SLRCLUB users according to the present invention.
12 is a graph showing rank correlation of SLRCLUB users according to the present invention.
13 is a graph showing the relationship between the rank of NAVER users according to the present invention in accordance with the present invention.
14 is a graph showing rank correlation of NAVER users according to the present invention.
15 is a graph illustrating cluster distribution of SLRCLUB and NAVER users according to the present invention.
16 is a graph showing rank correlation of users in a NAVER restaurant category and a travel category according to the present invention.
FIG. 17 is a graph illustrating a normalized discounted cumulative gain (NDCG) for a user's rank in SLRCLUB, NAVER travel category, and NAVER restaurant category according to the present invention.

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings. In the following description of the present invention, if it is determined that a detailed description of a known function or configuration may unnecessarily obscure the subject matter of the present invention, the detailed description thereof will be omitted. The following terms are defined in consideration of the functions of the present invention, and may be changed according to the intentions or customs of the user, the operator, and the like. Therefore, the definition should be based on the contents throughout this specification.

An online network is a space for communicating on the Internet, and includes various forms such as an online forum, an online community, a blogosphere, and a social network.

The reputation or expertise of a user on an online network, such as an online community, a blog, or a social network, is determined by online activities such as posts and feedback that the user has written online. For example, if another user provided feedback on a post written by a user, that user would have a higher reputation for that post and other users' feedback, and that user's score based on the reputation The score and rank will also rise.

According to an embodiment of the present invention, when a user obtains feedback from a user with a high reputation than a user receives feedback from a user with a low reputation, the user's reputation is increased. (Page Rank) algorithm or Hyperlink-Induced Topic Search (HITS) algorithm is applied.

1 is a graph model depicting feedback activity for a user's post. This graph model represents users 110, 120, 130 and posts 210, 220, 230, 240, 250 as nodes, respectively, and the user's post creation activity and feedback providing activity are user nodes and post nodes. Expressed as an edge connecting the. Specifically, the solid edge means the user's post writing activity, and the dashed edge means the user's feedback providing activity. For example, User 1 110 may have written Post 1 210 and Post 2 220 connected in solid lines and provided feedback to Post 3 230 and Post 4 240 connected in dotted lines. have.

2 is a graph model reconstructing the feedback activity of FIG. 1 in accordance with the present invention. In this graph model, only the users 110, 120, and 130 are represented as nodes, and the number of feedbacks 112, 121, 131, and 132 exchanged between users is represented as an edge connecting the user nodes. do. For example, user 1 110 may provide two feedbacks 112 to user 2 120 and receive one feedback 121 from user 2 120.

Each edge 112, 121, 131, and 132 of the graph reconstructed by the user center as shown in FIG. 2 represents an exchange of feedback between the corresponding users, and thus is suitable for calculating a user ranking based on reputation and calculating a search result accordingly. Contains information. Therefore, an embodiment of the present invention obtains a user ranking result by applying a page rank algorithm and an HITS-based algorithm to such a user-centered graph.

As described above, pagerank algorithms and HITS algorithms have been used to obtain the ranking of web pages and are being employed in commercial search engines. This algorithm may be applied to calculate the ranking of users in the network.

However, it can be seen that the characteristics of the links in the user-centered graph are quite different from those of the link between the user and the post. One difference is that in a user-centric graph, the edges between a pair of web pages are dichotomous. Users in an online network can provide one or more feedbacks to other users, so edges have weights. In the random walk model of the existing page rank model, the probability that a random surfer (or a user providing feedback) moves from one node to each of neighboring neighbor nodes is the same. However, when considering feedback, the user providing the feedback may then determine the user to provide feedback based on the frequency of feedback that the user previously provided to other users.

Another difference is the interaction between users. User interaction is an important characteristic in online networks. Online networks generally include a large number of intimacy (or friendship or acquaintance) groups. Users will give more feedback to users who belong to the same group than users who do not belong to the same group. This can be inferred by observing a very large number of feedback exchanges within a particular group. This tendency weakens the fairness of the ranking method based on the random walk model. For example, a user may have a higher ranking than the actual one that the user should receive because the user is very popular in the group to which the user belongs. In other cases, some users may have lower rankings than those that actually need to be because they are not in any familiar group. Therefore, in the case of using the existing ranking method, a user having a high social rank may have an unfairly high score, and thus a problem may occur in which a user having a low social rank has an unfairly low score.

Therefore, it is very important to consider sociality among users in calculating user ranking. In order to accurately measure the rank of each user, it is important to measure the sociality and reputation of each user. However, the sociality of the user does not always negatively affect the user's rank. Previous research has shown that the number of feedbacks exchanged with each other is deeply related to the rank of the user, such that the number of feedbacks exchanged between power users is greater than that of power users and non-power users. It is defined as the number of feedback exchanged between number> non-power users. The results of this conventional research are shown in FIG. 3.

3 is a graphical model illustrating the amount of interaction between users in accordance with the present invention. In FIG. 3, users are classified from level A to level E according to the amount of feedback between users. Power users include users at level A, and non-power users include users at level E. Most of the interactions occur between Level A users, and we can observe that there is little interaction between lower level users. Therefore, as described above, it can be recognized that sociality between users is important in calculating a user's rank.

Next, a method of calculating an initial reputation score, a method of calculating a social score, and a method of calculating a pure reputation score will be described in order to calculate a ranking of users in an online network.

* Initial Reputation Rank Rank ; RR )

The user's reputation rank may be calculated by applying the page rank algorithm and the HITS algorithm. As discussed above, edges are weighted because a user in an online network may provide one or more feedback to other users. Thus, in a user-centric graph, the edges between a pair of web pages are dichotomous. Therefore, in the existing page rank model, it is impossible to use the assumption that random surfers (or users providing feedback) have the same probability of moving from one node to each of neighboring neighbor nodes. Instead, the user providing the feedback may determine which user to provide feedback next based on the frequency of feedback that the user has previously provided to other users. The probability that the user v providing the feedback provides the next feedback to the user u _i is calculated as the ratio of the number of feedbacks that v provides to u _i and the number of feedbacks that v provides to the entire user.

Therefore, based on the random walk model, the user's initial reputation score RR may be calculated as in the following equation.

[Equation 1]

는 v가 작성한 피드백의 전체 개수, d는 감쇄 인자(damping factor)를 나타낸다. 즉, 최초 평판도 점수는 u에게 다른 사용자들이 피드백을 제공할 확률과 다른 사용자들의 최초 평판도 점수의 곱셈에 의하여 계산된다.Where RR _rw (u) is u's initial reputation score, RR _rw (v) is v's initial reputation score, b (u) is the set of users who provided feedback to u, and f (v) is feedback written by v The set of users who received F _{v- > u} is the number of feedback v has given to u,

Is the total number of feedbacks written by v, and d is the damping factor. That is, the initial reputation score is calculated by multiplying the probability that other users give feedback to u and the initial reputation score of other users.

On the other hand, based on the HITS algorithm, the user's initial reputation score (RR) can be calculated as shown below.

[Equation 2]

In Equation 2, RR _{hits - authority} represents a user's score based on authority, and RR _{hits - hub} represents a user's score based on butt. In the case of applying the HITS model, the user v is defined as the hub node, the user u is defined as the authority node, and the authority score of u is calculated, and the initial reputation score of u is calculated according to the authority score of u. In this case, the authority score of u may be calculated by calculating the hub score of v according to the initial authority score of u, and calculating the product of the probability that v provides feedback to u and the hub score of v. The user's initial reputation score, RR _hits , based on the HITS algorithm, has a value similar to RR _{hits-authority} .

Meanwhile, in order to calculate a more accurate user ranking as described above, in addition to the initial reputation score, the social score between users of the online network may be further considered. Hereinafter, a method of calculating social rank between online network users will be described.

Sociability Rank (SR)

In the online network, the relationship between users can be classified into two types. One is a relationship created by friendship and the other is a relationship created by reputation. Therefore, in order to efficiently obtain user ranking results, it is necessary to classify the relationship between users on the online network into a social relationship and a reputation relationship.

4 is a graph model classifying relationships between online network users into sociality and reputational relationships according to the present invention. In FIG. 4 (a), both social and reputation relationships between users are illustrated, which makes it difficult to calculate accurate user rankings. On the other hand, Figure 4 (b) shows the social relationship formed by the friendship between users. 4 (c) shows the flatness relationship formed only by the flatness between users. Therefore, it is recognized that there is a social relationship and a reputation relationship among online network users, and a method of calculating a social rank formed by the social relationship will be described below.

Sociality is an important factor that affects the ranking of users in online networks. If the two users form an intimate relationship with each other, there is a high possibility of exchanging feedback compared to the case where the two users do not form an intimate relationship with each other. In addition, it is more likely that users with high impact exchange feedback than users with low impact. Sociality is calculated taking into account the bidirectionality of feedback between users.

FIG. 5 is a graphical model illustrating social strength between users in accordance with the present invention. FIG. When two users on the online network are referred to as user a and user b, a post written by user a may be referred to as P _a 510 and a post written by user b may be referred to as P _b 520. On the other hand, the number of feedbacks provided by a to a post written by _b is referred to as F _{a- > b} 530, and the number of feedbacks provided by b to a post written by a is F _{b- > a} 540. Is referred to.

On the other hand, the social strength between users a and b is referred to as F ^(s) _{a <-> b} 550. _{^{F (s) a <->}} b is F is determined on the basis of the value _{of> a - 'a -> b,} and F' _b. Where, F _{'a -> b,} and F' _{b -> a} are each F _{a -} may _{be> a (540) -> b} (530) and F _b. In addition, F ' _{a- > b} and F' _{b- > a} may be calculated as in Equation 3 below.

[Equation 3]

That is, F ' _{a- > b} and F' _{b- > a} may be calculated in consideration of the number of posts created by a and b. In addition, F ' _{a- > b} and F' _{b- > a} may be calculated as in Equation 4 below.

[Equation 4]

After F ' _{a- > b} and F' _{b- > a} are generated, F ^(s) _{a <-> b} is calculated. _{^{F (s) a <->}} b is F _{'a -> b,} and F' _{b - ->} smaller of a, F _{'a -> b,} and F' _{b -> a} of the arithmetic mean, the geometric mean and the harmonic mean Can be calculated to either. _{^{F (s) a <->}} b a F _{'a -> b,} and F' _{b -} when calculating the harmonic mean _{of> a} formula shown in the following formula 5.

[Equation 5]

If a and b do not give each other feedback, the social strength F ^(s) _{a <-> b} (550) is zero. The social score (SR) of the user may be calculated using the social strength F ^(s) _{a <-> b} 550 calculated above. The social score is calculated as shown in Equation 6 below.

[Equation 6]

는 v와 다른 사용자간의 사회성 점수의 합, d는 감쇄 인자를 나타낸다. 수식 6은 랜덤 워크 모델을 적용하여 계산한 것이나, 사회성 강도를 HITS 모델에 적용하여 사회성 점수를 계산할 수도 있다. 수식 6에서 계산된 사회성 점수는 후에 순수 평판도 점수를 계산함에 있어서 고려될 수 있다. Where u is the first user, v is the second user, SR (u) is the social score of u, SR (v) is the social score of v, bd (u) is the set of users who exchanged u feedback, bd ( v) is the set of users who exchanged feedback with v, F ^(s) _{v-> u} is the social score of u and v,

Is the sum of the social scores between v and other users, and d is the decay factor. Equation 6 is calculated by applying a random walk model, but may also calculate a social score by applying the social strength to the HITS model. The social score calculated in Equation 6 may be considered later in calculating the pure reputation score.

Hereinafter, a method of calculating a pure reputation score between online network users by using the aforementioned initial reputation score and social score will be described.

Pure reputation Rank  ( Pure Reputation Rank ; PR )

The pure reputation score can be obtained by subtracting the sociality score from the original reputation score. For example, if a user's reputation is very high, if the majority of the user's scores are social scores, it is difficult to see this ranking as an accurate ranking. Therefore, in order to accurately calculate the user's ranking score, it is very important to separate the user's social score and the user's pure reputation score.

6 is a graph model showing the pure plate strength between users in accordance with the present invention. User a and b are exchanging feedback on the online network, and the post written by user a is P _a 610 and the post written by user b is referred to as P _b 620. In addition, the number of feedbacks provided by a to the post written by _b is referred to as F _{a- > b} 630, and the number of feedbacks provided by b to the post written by a is F _{b- > a} (640). Is referred to. Also, the social strength between users a and _b is referred to as F ^(s) _{a <-> b} .

In this case, F ^(p) _{a-> b} 650, which is the pure ^plate strength from user a to b, may be calculated as in Equation 7 below.

[Formula 7]

Basically, pure flatness strength to b from a user and a is the number of feedback to b from a user _{a (F a -> b)} - can be subtracted to calculate the user social between a and b, the intensity _{^{(> b F (s) a}} <) in have. However, F _{a ->} in _b F ^(s) _{a <- because>} if subtracting the _b may come out a value less than 0 F ^(p) _min generate is value, F ^(p) _min and F _{a -> b} -F ^{(s) The} larger of _{a <-> b} is calculated as the pure plate strength from a to b. F ^(p) _min may be set to a smaller value of _a predetermined constant f _min and F _{a- > b} .

For _{example, F a -> b = 1} , F b -> a = 3, F (s) a <-> b = 1.5, when the ^{_{f min = 0.5, F (p}} ) a-> b = 0.5 of With a value of F ^(p) _{b-> a} = 1.5. In calculating the net reputation strengths of a and b, the net reputation strengths of a and b are lower than the original reputation because the sociality between a and b is taken into account.

The pure reputation score of the user may be calculated by applying the aforementioned pure reputation intensity to a random walk model. The pure reputation score (PR) of the user is expressed as shown in Equation 8 below.

[Equation 8]

는 u로부터 v로의 순수 평판도 강도,

는 v의 순수 평판도 강도의 합, d는 감쇄 인자(damping factor)를 나타낸다. 한편, HITS 모델을 적용하여 사용자의 순수 평판도 점수를 계산할 수도 있다. Where u is the first user, v is the second user, PRrw (u) is the pure reputation score of u, PRrw (v) is the pure reputation score of v, and b (u) is the set of users who provided feedback to u , f (v) is the set of users for whom v provided feedback,

Is the pure platen strength from u to v,

Is the sum of the pure platelet strengths of v, d is the damping factor. Meanwhile, the pure reputation score of the user may be calculated by applying the HITS model.

7 is a flowchart illustrating a schematic flow of a reputation calculation method of an online network user implemented in a reputation calculation apparatus according to the present invention. The reputation calculator includes a receiver, a first information acquisition unit, an initial reputation score calculation unit, a second information acquisition unit, a sociality score calculator, and a pure reputation score calculator. In step S710, the receiving unit receives feedback information on a plurality of online network posts written by a plurality of online network users. Feedback may include one or more of a group that includes comments, comments, empathy, recommendations, trackbacks, scraps, and links.

In operation S720, the first information acquisition unit acquires first information that is a probability that each of the users who are not the first user among the users provides feedback to the first user. The first information may be obtained by dividing the number of feedbacks provided by the second user among the users to the first user by the total number of feedbacks provided by the second user to other users.

In operation S730, the first reputation score calculator calculates an initial reputation score of the first user using the first information. In an online network, the more a user's posts receive feedback from a user with a higher reputation, the higher the reputation (compared to a feedback from a user with a lower reputation). Therefore, in order to calculate the first reputation score of the first user, the multiplication of the first information and the first reputation degree of the second user is added and calculated. In addition, the first reputation score of the first user may be calculated by Equation 1 applying the random walk model and Equation 2 applying the HITS model.

In operation S740, the second information acquisition unit acquires second information, which is a probability that each of the first user and the entire user exchanges feedback. The second information is calculated based on social strength, and is obtained by dividing the number of feedbacks (social strength) exchanged with the second user by the first user by the total number of feedbacks exchanged with other users by the second user. In this case, the number of feedbacks exchanged with the second user by the first user is the number of first feedbacks using the number of feedbacks provided by the first user to the second user and the number of feedbacks provided by the second user to the first user. It is calculated using the second feedback number using. Here, the number of feedbacks exchanged with the second user by the first user is calculated as one of a smaller value of the first and second feedback numbers, an arithmetic mean, a geometric mean, and a harmonic mean of the first and second feedback numbers. Can be. The first feedback number and the second feedback number may be calculated in consideration of the number of posts written by the first user and the number of posts written by the second user.

In operation S750, the social score calculator calculates the social score of the first user using the second information. In this case, as in the case of calculating the first reputation score of the first user, the social walk score of the first user may be calculated by applying a random walk model, including the multiplication of the second information and the social score of the second user. The social score of the first user may be calculated by Equation 6.

In operation S760, the pure reputation score calculator calculates a pure reputation score of the first user using the first reputation score and the sociality score of the first user. In this case, the net reputation score of the first user may be calculated by subtracting the first social user score from the first reputation score of the first user. In addition, in case a value obtained by subtracting the social user score of the first user from the initial reputation score of the first user is 0 or less, a predetermined threshold value and a value obtained by subtracting the social user score of the first user from the initial reputation score of the user. A large value can be calculated as the net reputation strength of the first user for the second user. The process of calculating the pure reputation score by applying the pure reputation intensity to the random walk model is shown in Equation 8.

Meanwhile, the final reputation score may be calculated using the pure reputation score and the sociality score. The final reputation score is calculated by Equation 9.

[Equation 9]

At this time, a is set to a constant between 01.

8 is a block diagram illustrating a schematic configuration of an apparatus for calculating a reputation of a network user according to the present invention. The receiver 800 receives feedback information about a plurality of online network posts written by a plurality of online network users. The first information acquisition unit 810 calculates a probability that each of the users other than the first user among the users provides feedback to the first user. The first reputation score calculator 820 calculates the first reputation score of the first user using the first information acquired by the first information acquirer 810. The second information acquisition unit 830 obtains second information that is a probability that the first user and each of the users exchange feedback. The social score calculator 840 calculates the social score of the first user using the second information acquired by the second information acquirer 830. Meanwhile, the pure reputation score calculator 850 calculates the pure reputation score of the first user using the initial reputation score calculated by the initial reputation score calculator 820 and the social score calculated by the social score calculator 840. do.

Meanwhile, the pure reputation score calculator 850 may further include a subtraction unit 860 that subtracts the social score of the first user from the initial reputation score of the first user. Also, the reputation calculator may include a final reputation score calculator 870, and the final reputation score calculator 870 may be a social score and a pure reputation score calculator 850 calculated by the social score score 840. The final reputation score of the first user is calculated using the pure reputation score calculated in.

Hereinafter, an experimental result of calculating a user's reputation using the network user's reputation method and apparatus will be described.

Experiment result

Two data sets were collected for the experiment. One is SLRCLUB ( http://slrclub.com ), an online photo sharing site like Flickr. 147,387 posts and 1,883,239 feedback by 144,514 users were collected. Another data set is the NAVER blog ( http://blog.naver.com ), which collected 4,382,386 blog posts and 7,642,440 feedback by 1,107,208 users. Most of these data were generated in 2008. Table 1 shows the statistics of the collected data.

[Table 1]

Hereinafter, an initial reputation score will be expressed as an RR score, a sociality score as an SR score, and a pure reputation score as a PR score. On the other hand, the first reputation rank is represented by RR rank, the social rank is represented by SR rank, and the pure reputation rank by PR rank.

9 is a graph showing a SLRCLUB user score distribution according to the present invention. Figure 9 (a) is a graph showing the RR score according to the RR rank, Figure 9 (b) is a graph showing the PR score according to the PR rank, Figure 9 (c) is a graph showing the SR score according to the SR rank. . Of the 144,514 users of SLRCLUB, only 40.000 users have a non-zero user score. On the other hand, it can be seen that only 5,000 users have an SR score other than zero. As a result, only a very small number of people interact with each other and the PR score is calculated by subtracting the SR score. Therefore, the PR score is generally smaller than the RR score. The SR score has a value much smaller than the PR score or the RR score.

10 is a graph illustrating a relationship between ranks of SLRCLUB users according to the present invention. Fig. 10 (a) is a graph showing users with SR rank as x axis and PR rank as y axis, and Fig. 10 (b) is a graph showing users with SR rank as x axis and RR rank as y axis. to be. 10 (a) and 10 (b) are calculated using the page rank algorithm. Comparing Figs. 10 (a) and 10 (b), it can be seen that 10 (b) has less data in the vicinity of the x = y line. That is, it shows that PR rank does not depend on SR rank compared with RR rank. 10 (c) and 10 (d) are graphs showing the relationship between SR rank-RR rank and SR rank-PR rank, respectively, using the HITS algorithm, similarly to 10 (a) and 10 (b). It can be seen that the PR rank does not depend on the SR rank compared to the RR rank.

11 is a graph illustrating a user having a high rank among SLRCLUB users according to the present invention. User 1 has a RR ranking of 30, but has a low SR ranking (274, etc.). Therefore, if the PR ranking is calculated in consideration of the sociality of user 1, it can be seen that it is 1, and so on. On the other hand, the user 2 has only RR ranking of 2 and PR ranking of 12 and the like. Since user 2 has an active sociality such as SR ranking of 2, PR ranking that considers reputation only except sociality becomes low. In addition, the user 3 can be found that the RR ranking is 1, the SR ranking is 1, and the PR ranking in consideration of sociality is 4th, and the PR ranking is not much lower than the RR ranking.

12 is a graph showing rank correlation of SLRCLUB users according to the present invention. It can be seen that the correlation between the SR rank and the RR rank has an average of about 85% correlation. On the other hand, the correlation between the SR rank and the PR rank shows a relatively low distribution. In particular, the top 30 users have a correlation of 16%, and as a whole, there is a 53% correlation, indicating that the value is lower than the correlation between the SR rank and the RR rank. This indicates that the pure reputation score of the user is a value in which the social score is successfully separated from the initial reputation score of the user.

13 is a graph illustrating a relationship between ranks of NAVER users according to the present invention. 13 (a) and 13 (b) are graphs showing the relationship between SR rank and RR rank and the SR rank and PR rank, respectively, using a page rank algorithm. 13 (c) and 13 (d) are graphs showing the relationship between SR rank and RR rank and the SR rank and PR rank, respectively, using the HITS algorithm. It can be also seen from FIG. 13 that the PR rank does not depend on the SR rank as compared to the RR rank.

14 is a graph showing rank correlation of NAVER users according to the present invention. However, unlike in FIG. 12, it can be seen that the correlation between the SR rank and the RR rank is low. In particular, this phenomenon can be seen more prominent among the top rankers. This difference is due to the nature of the network. Unlike photo-sharing networks such as SLRCLUB, the NAVER blog space has a variety of topics, and users tend to be interested in only a few topics. Therefore, the total number of users' SRs is generally low, and the correlation between the SR rank and the RR rank is also low. The trend among these users is shown in the cluster distribution in FIG. 15.

15 is a graph illustrating cluster distribution of SLRCLUB and NAVER users according to the present invention. The users of SLRCLUB tend to be concentrated around the emphasis, but the users of NAVER blogs are scattered due to the wide variety of NAVER blog topics and narrow personal interests. To further observe this trend, we classified users into two categories and observed their characteristics within the NAVER blog. The two categories are the restaurant category and the travel category, and about 110 users were extracted from the restaurant category and about 200 users were extracted from the travel category.

16 is a graph showing rank correlation of users in a NAVER restaurant category and a travel category according to the present invention. 16 (a) and 16 (b) are graphs showing the relationship between SR rank and RR rank and the SR rank and PR rank, respectively, using a page rank algorithm. 16 (c) and 16 (d) are graphs showing the relationship between SR rank and RR rank and the SR rank and PR rank, respectively, using the HITS algorithm. It can also be seen from FIG. 16 that the PR rank does not depend on the SR rank as compared to the RR rank.

Table 2 below shows the correlation between the SR rank and the RR rank of the NAVER restaurant category and the travel category, the correlation between the SR rank and the PR rank, and the correlation between the RR rank and the PR rank.

[Table 2]

Since the difference between the SR-RR correlation and the SR-PR correlation in the restaurant category is larger than the difference in the travel category, the RR-PR correlation in the restaurant category has a lower value than the RR-PR correlation in the travel category. do.

On the other hand, Normalized Discounted Cumulative Gain (NDCG) is an evaluation metric that measures the rank accuracy of search results.It extracts the top 20 users from three data sets, including two categories, SLRCLUB and NAVER, and then targets 10 testers. Each user was given a score.

FIG. 17 is a graph illustrating a normalized discounted cumulative gain (NDCG) for a user's rank in SLRCLUB, NAVER travel category, and NAVER restaurant category according to the present invention. In each graph, it can be seen that PR has higher NDCG than other ranking algorithms.

As a result of the above experiment, it can be seen that the reputation score of online network users is largely due to sociality among users. In addition, it was found that separating sociality is an effective way to obtain accurate user scores.

Modules, functional blocks or means of the present embodiment may be implemented in a variety of known elements, such as electronic circuits, integrated circuits, ASICs (Application Specific Integrated Circuit), each may be implemented separately, or two or more may be integrated into one Can be.

Although the embodiments have been described for the understanding of the present invention as described above, it will be understood by those skilled in the art, the present invention is not limited to the specific embodiments described herein, but variously without departing from the scope of the present invention. May be modified, changed and replaced. Therefore, it is intended that the present invention cover all modifications and variations that fall within the true spirit and scope of the present invention.

800: receiving unit 810: first information obtaining unit
820: First reputation score calculator 830: Second information acquisition unit
840: social grade score calculator 850: pure reputation score calculator
860: subtraction unit 870: final reputation score calculation unit

Claims (3)

As a reputation calculation method of an online network user implemented in a reputation calculation device,
The flatness calculation device,
Receiving feedback information on an online network post written by a plurality of online network users at a receiving unit;
Acquiring, by a first information acquisition unit, first information that is a probability that each of the users provides feedback to a first user of the users;
Calculating an initial reputation score of the first user by using the first information in an initial reputation score calculator;
Acquiring, by a second information obtaining unit, second information which is a probability that each of the first user and the user exchanges feedback;
Calculating a sociability score of the first user by using the second information in a social score calculator; And
Calculating a pure reputation score of the first user by using the first reputation score and the sociality score of the first user in the pure reputation score calculator;
How to calculate the reputation of online network users.
A receiver configured to receive feedback information on an online network post written by a plurality of online network users;
A first information obtaining unit obtaining first information that is a probability that each of the users provides feedback to a first user of the users; And
And a reputation score calculator configured to calculate a reputation score of the first user using the first information.
Reputation calculator for online network users. A receiver configured to receive feedback information on an online network post written by a plurality of online network users;
A first information obtaining unit obtaining first information that is a probability that each of the users provides feedback to a first user of the users;
An initial reputation score score calculator for calculating an initial reputation score of the first user using the first information;
A second information obtaining unit obtaining second information that is a probability that each of the first user and the user exchanges feedback;
A social score calculator that calculates a social score of the first user using the second information; And
It includes a pure reputation score calculation unit for calculating the pure reputation score of the first user using the first reputation score and sociality score of the first user
Reputation calculator for online network users.

Images