KR100872305B1 - Method and system for analyzing blog - Google Patents

Abstract

The present invention relates to a blog analysis method and system, the method comprising: calculating user content power of a plurality of first users, calculating user delivery power of a plurality of second users, and user content power and user Calculating a propagation probability that the posting of the first user is spread to the second user based on the transmission power. According to the present invention, a power user who is central to establishing various policies for activation of a blog service is calculated by calculating a propagation probability based on the user content power and the user delivery power of each user in a blog network and applying it to an independent propagation model. Can be selected.

Blog, Post Content Power, User Content Power, Power User, Independent Propagation Model, Ripple Probability

Description

Blog Analytics Method and System {METHOD AND SYSTEM FOR ANALYZING BLOG}

1 is a block diagram illustrating a blog analysis system according to an exemplary embodiment of the present invention.

2 is a diagram illustrating a blog network.

3 is a block diagram of the blog analysis server shown in FIG. 1.

4 is a flowchart illustrating a blog analysis method according to an embodiment of the present invention.

5 is a diagram illustrating an example of calculating a post content power according to an embodiment of the present invention.

FIG. 6 is a view illustrating exposure time and exposure time correction values of a post to explain an example of calculating user content power according to an embodiment of the present invention.

7 is a schematic diagram illustrating an independent propagation model used in a blog analysis method according to an exemplary embodiment of the present invention.

8 is a schematic diagram illustrating an interpolation method for determining user content power weight and user delivery power weight according to an embodiment of the present invention.

9 is a schematic view for explaining the response and precision used in the blog analysis method according to an embodiment of the present invention.

10 is a view showing an example of the ripple relationship between blog users using the ripple history in accordance with an embodiment of the present invention.

FIG. 11 is a diagram illustrating a blog connection network in which the ripple relationship between users illustrated in FIG. 10 is displayed.

<Description of Drawing>

100: blog server, 130: blog analytics server,

132: ripple history generation unit, 134: article content power calculation unit,

136: user content power calculator, 137: user transfer power calculator,

138: ripple probability calculation unit, 139: power user selection unit,

160: database, 200: communication network,

300: user terminal

The present invention relates to a blog analysis method and system.

A social network is a network that expresses the relationship structure of a society as a whole by using relationship information between members in a society. It is called social network analysis to study the relationship structure and derive the unique characteristics of the society. In the past, it was difficult to obtain data indicating the relationship between members in social networks. Therefore, the focus of the existing research was to model the structure and characteristics of social networks without specific data on the relationship between members.

Due to the development of the Internet, these social networks have appeared online. The most distinctive feature of the online social network is that, unlike the existing social network, it stores specific information that can explain the relationship between members and information exchanged by the relationship in a database. Due to these characteristics, recent researches have been conducted to analyze online social networks.

A blog is a kind of personal website where users can store their posts online, and the blog world is a representative online social network of such blogs. Each blog user can have a relationship with someone else's blog, and these relationships form an online social network. Hereinafter, a social network consisting of a relationship between a blog and a blog is called a blog network.

The goal of an operator or company that runs a blogging world is to increase the profitability of the blogging world as a whole. Within the blog network, there are special blog users who play a role in encouraging other general blog users to actively use the blog service. If you can identify these users, you will be able to formulate various policies around blog service activation.

However, according to the previous researches, only the topological features of the social network have been considered to identify these users. However, if the criteria using only the topological features of the social network are used in this way, it is impossible to correctly select the users who have a great influence on the actual network. For example, a user with many neighbors in a blog network is not necessarily more influential to other users. Rather, a user with a few neighbors can exert a greater influence on other users. Therefore, in this case, it is difficult to identify blog users who have a substantial influence on other users.

In addition, according to the blog network considering only topological features, it is difficult to grasp how the blog network is structurally connected according to user's actions, and it is difficult to grasp the ripple relationship between users.

Therefore, the technical problem to be achieved by the present invention is to provide a blog analysis method and system that can select a blog user that can exert a substantial influence on other users and is central to establishing various policies for activation of a blog service.

In addition, another technical problem to be achieved by the present invention is to provide a blog analysis method and system that can structurally connect each blog in the blog network, and can properly grasp the ripple relationship between blog users.

According to one or more exemplary embodiments, a blog analysis method includes calculating user content powers of a plurality of first users, calculating user delivery powers of a plurality of second users, and Calculating a propagation probability that the posting of the first user spreads to the second user based on user content power and the user delivery power.

The user delivery power may be calculated by multiplying the scrap weight and the chained weight by the number of times that the second user scraps and posts the other user's posts, and then adds the resulting value.

The propagation probability may be calculated by multiplying the user content power weight and the user delivery power weight by a value obtained by normalizing the user content power and the user delivery power between 0 and 1, respectively, and adding the resultant value.

The user content power is normalized by dividing the value of the user content power corresponding to the top x% by setting the upper x% or more of the user content power to be analyzed as 1 and the user delivery power is the upper y% of the user delivery power to be analyzed. With the above value set to 1, it can be normalized by dividing by the value of the user transmission power corresponding to the upper y%.

The user content power weight and the user transfer power weight may be determined by interpolation so that the ripple prediction accuracy according to the ripple probability is maximized.

The ripple prediction accuracy may be calculated as an average of responsiveness and precision between a user who actually spreads a predetermined article and a user who is predicted to spread through the independent propagation model.

The method may further include selecting a power user set based on the propagation probability and the independent propagation model.

The step of selecting a power user set may include calculating the propagation power of the first user by applying the propagation probability to the independent propagation model, selecting a third user whose maximum propagation power is maximum, and the third The method may include repeating a predetermined number of the stepped power calculation step and the third user selection step except for a user who is expected to be assimilated by the user.

The calculating of the user content power may include calculating a post content power of the post based on an action frequency of a blog user for at least one post, and correcting the post content power according to an exposure time of the post. And the user content power may be calculated by adding the corrected post content power.

A computer readable medium according to another aspect of the present invention records a program for causing a computer to execute any of the above methods.

According to another aspect of the present invention, a blog analysis system includes: a user content power calculating unit calculating user content power of a plurality of first users, a user delivery power calculating unit calculating user delivery power of a plurality of second users, and the user content And a propagation probability calculator configured to calculate a propagation probability that the post of the first user spreads to the second user based on power and the user transmission power.

The user transfer power calculator may calculate the user transfer power by multiplying the scrap weight and the chained weight by the number of times the second user scraps and posts the other user's posts, and adds the result value. have.

The propagation probability calculator may calculate the propagation probability by multiplying the user content power weight and the user delivery power weight by a value normalized between the user content power and the user delivery power to 0 and 1, respectively, and adding the resultant value.

The user content power calculation unit sets the upper x% or more of the user content power to be analyzed as 1, and normalizes the user content power by dividing the user content power by a value of the user content power corresponding to the upper x%. The upper y% or more of the target user delivery power may be set to 1, and the user delivery power may be normalized by dividing by the value of the user delivering power corresponding to the upper y%.

The propagation probability calculator may determine the user content power weight and the user transfer power weight by interpolation so that the propagation prediction accuracy according to the propagation probability is maximized.

The propagation probability calculator may calculate the propagation prediction accuracy based on an average of responsiveness and precision between a user who actually propagates a predetermined post and a user predicted to be propagated through an independent propagation model.

The apparatus may further include a power user selecting unit configured to select a power user set based on the propagation probability and the independent propagation model.

The power user set selecting unit applies the propagation probability to the independent propagation model to calculate the propagation power of the first user, selects a third user whose maximum propagation power is maximum, and is animated by the third user. A predetermined number of operations of calculating the propagation power and selecting the third user may be repeated for a user other than a user who is expected to return.

And a post content power calculator configured to calculate a post content power of the post based on a blog user's action frequency with respect to at least one post, wherein the user content power calculator determines the post content power according to the exposure time of the post. The user content power may be calculated by adding the corrected post content power.

DETAILED DESCRIPTION Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings so that those skilled in the art may easily implement the present invention.

First, referring to FIGS. 1 to 3, a blog analysis system and a blog network to be analyzed in accordance with an embodiment of the present invention will be described in detail.

1 is a block diagram illustrating a blog analysis system according to an embodiment of the present invention, FIG. 2 is a diagram illustrating a blog connection network, and FIG. 3 is a block diagram of the blog analysis server shown in FIG. 1.

Referring to FIG. 1, a blog analysis system according to an embodiment of the present invention includes a blog server 100, a blog analysis server 130, and a database 160, and includes a plurality of user terminals through a communication network 200 ( 300).

The blog server 100 provides a blog service to a user. A blog is a kind of personal web site where users can post freely according to their interests. The blog server 100 provides a blog to the user and allows the user to create a document (or post) including text, images, videos, music, and the like. Make sure that you write to your blog. In addition, the blog server 100 transmits and displays its own blog document and another user's blog document to the corresponding user terminal 300 through the communication network 200 at the request of the user.

The communication network 200 may include a telephone network as well as a data communication network including a local area network (LAN), a metropolitan area network (MAN), a wide area network (WAN), the Internet, and the like. It can be used, and it does not matter wired and wireless, and it does not matter which communication method is used.

The user terminal 300 connects to and receives information from the blog server 100 and the blog analysis server 130 through the communication network 200, and displays the blog document from the blog server 100 in a browser (not shown). do. The user terminal 300 is provided with memory means such as a laptop computer, a workstation, a palmtop computer, a personal digital assistant (PDA), a web pad, a mobile communication terminal, as well as a desktop computer, and includes a microprocessor. It can be made of a terminal equipped with a computing power.

Each user may write a post in his blog through the user terminal 300. The other user can query, leave a comment, scrap, post a reply, etc. on the post through his user terminal 300. Here, inquiring is an action to read the post, leaving a comment is an action to leave their comments about the post, scraping is an action to copy the content of the post and register as a post in their blog, post a post This action creates a new post in your blog. Like normal posts, posts reproduced through scraping and posting can also trigger actions such as viewing, commenting, scraping, and posting a post by another user.

Referring to FIG. 2, the thick squares represent the blogs BA, BB, BC, and BD corresponding to the users U _A , U _B , U _C , and U _D , respectively, and the small squares inside the posts represent the posts existing in the blog. . The arrow shows the relationship between article 1 of user U _A and posts reproduced by the actions taken by users U _B , U _C , and U _D. For user U _A 's post 1, other users took various actions such as viewing, commenting, scraping, and posting tracks. In particular, user U _B scraped this post as his second post, and user U _C registered it as his third post by posting a reply to this post. In addition, user U _D scraped post 3 of user U _C as his second post.

The blog server 100 records the actions that the user has made about his blog and other users' blogs. In other words, the blog server 100 is information about all the situations that occur in the blog, for example, the history of the user's posts on his blog, the history of the user viewed the blog posts of other users, the history of scraping, Blog data including a history of commenting on another user's blog post, a user's history of writing another user's blog post, and the like are recorded in the database 160.

The database 160 receives and stores blog posts and blog data from the blog server 100, and provides corresponding posts and data according to requests from the blog server 100 and the blog analysis server 130. Depending on the type of blog document, text, images, videos, etc. can be stored in a separate database, and blog data, etc. can also be stored in a separate database.

Referring to FIG. 3, the blog analysis server 130 may include a spread history generator 132, a post content power calculator 134, a user content power calculator 136, a user transfer power calculator 137, and a spread probability calculator 138. And a power user selector 139.

The spread history generator 132 generates a spread history based on blog data from the database 160. Ripple history is a set of data rearranged in a structure that allows you to effectively search for ripple relationships between blog posts. The post content power calculator 134 calculates the frequency of the user's action for each post with reference to the spread history, and calculates the content power of each post based on the frequency and weight of the action. The user content power calculator 136 calculates the user content power based on the content power of all posts owned by each user. The user transfer power calculator 137 calculates the transfer power of each user with reference to the propagation history. The delivery power here indicates how active the user is in reproducing the content of other users. The spread probability calculator 138 calculates the spread probability between users in the blog network based on the user content power and the user delivery power. The power user selection unit 139 selects a power user by applying the calculated propagation probability to an independent cascade model. A power user is a user who spreads content to many other users in a blog network.

The blog analysis server 130 may be implemented by being included in the blog server 100. The blog analysis server 130 may be implemented as a separate system without being connected to the blog server 100 and the blog database 160 to receive blog data and perform a corresponding function. You may. At least one of the propagation history generator 132, each of the calculators 134, 136, 137, and 138, and the power user selector 139 may be implemented separately from the blog analysis server 130 as a separate module.

4 will be described in detail with reference to the blog analysis method according to an embodiment of the present invention.

4 is a flowchart illustrating a blog analysis method according to an embodiment of the present invention.

First, terms and symbols are defined for future description. U _i means a user whose identifier is i. D _i means a set of posts owned by U _i , and D _{i, j} means the j th post of U _i . The influence of the content of the article D _{i, j} on other users is defined as the document contents power of this post, and is expressed as DCP (D _{i, j} ). In addition, the influence of the contents of the user U _i on other users is defined as the user contents power, and is expressed as UCP (U _i ). As mentioned earlier, there are five actions a user can take using the blog service: write, read, comment, scrap, and link. These actions are denoted by W, R, C, S, and L, respectively. When quantifying the influence of the content of a post, different weights can be assigned to give different actions to each action. The weights for actions W, R, C, S, and L are denoted by ww, wr, wc, ws1, wl1.

Referring to FIG. 4, the propagation history generator 132 generates a propagation history based on blog data from the database 160 (S410). The spread history is a structural representation of the history of the spread of this original post based on the original original post.

Table 1 shows an example of the ripple history generated for one original post. The first line of the propagation history shows information about the original post, and the lines below show information about the post created based on this original post. In this case, UID is a user identifier, D # represents an identifier of a corresponding post registered in each user blog, and TYPE indicates through which action the post is registered. PUID and PD # represent the user identifier of the last post and the identifier of the last post, respectively, which are the source of the post's ripple, and TIME means the time when the post was registered.

Referring to [Table 1], the user U _A has created and registered the original post D _{A, 1} at t1 (W), and the user U _C has posted the post (L) at t2 for the original post D _{A, 1} . Has registered the article D _{C, 3} . In addition, user U _B scraped the original post D _{A, 1} in t3 (S) _, and registered the post D _{B, 2} , and user U _D scraped the post D _{C, 3} registered by the user U _C to t4 (S) Post D _{D, 2} has been registered.

In this way, by creating a ripple history for each original post, it is easy to check the ripple relationship between the posts can effectively calculate the power of the post content.

The post content power calculating unit 134 calculates direct and indirect content power of each post based on the generated spill history (S420 and S430). If another user shows an action on a particular post, that means that the post was affected by the other user. Therefore, by multiplying the frequency of each action of the user for each post and the corresponding weight, the content power of the post can be quantified.

Each post can affect other users in the first blog you create, and it can affect other users in other blogs that have been propagated through scrapping or posting. The former is called the direct effect of the post and the latter is called the indirect effect of the post, and the quantified values are defined as the direct contents power and the indirect contents power of the post respectively. Posts D _i, referred to direct content-power and indirect contents power of _j DRCP (D _{i, j)} and (D _{i, j)} IDCP when each title, as an example, they Equation 1 and Equation 2 It can be calculated as Of course they can be calculated in different ways.

DRCP (D _{i, j} ) = wr × RC (D _{i, j} ) + wc × CC (D _{i, j} ) + ws1 × SC (D _{i, j} ) + wl1 × LC (D _{i, j} )

Here, RC (D _{i, j} ), CC (D _{i, j} ), SC (D _{i, j} ), LC (D _{i, j} ) refer to the number of times the article D _{i, j} has been viewed, and the article D _{i, j} comments left on the count means the number of trackbacks only for the posts D _i, a number of scrap _j, posts D _{i, j.}

Here, the post D _{i ', j'} represents a post reproduced directly from the post D _{i, j} , and wd and wi are direct content power weights and indirect content power weights, respectively.

The post content power calculator 134 calculates the content power of each post based on the calculated direct and indirect content power (S440). The post content power can be calculated as the sum of the direct content power and the indirect content power of the post as shown in [Equation 3]. I can regulate it.

DCP (D _{i, j} ) = wd × DRCP (D _{i, j} ) + wi × IDCP (D _{i, j} )

5, an example of calculating content power of a post using a spill history will be described. FIG. 5 is a diagram illustrating an example of calculating a post content power according to an exemplary embodiment of the present invention, and illustrates an example of calculating the content power of a post by using a ripple relationship between the posts shown in the propagation history of [Table 1]. . Here, the weight wd of the direct content power and the weight wi of the indirect content power are equally set to 1, respectively. The value in each post in turn means (Direct Content Power, Indirect Content Power, Post Content Power). Posts D _{D, 2} have no post-repeat record, so the direct content power of 5 becomes the post content power. Therefore _, the indirect content power of D _{C, 3} is 5, and the post content power of D _{C, 3} is 8 by adding 3, which is a direct content power. In addition, D _{A, 1,} which is the original source of the post _, has 12 as the indirect content power by adding the post content power of D _{C, 3} , 8, and the post content power of D _{B, 2} , 4 to 12. Add 4 to D _{A, 1 to have} a post content power of 16.

Referring back to FIG. 4, the user content power calculator 136 calculates the content power of each post and calculates the user content power using the post content at step S450.

On the other hand, the content power of the post tends to increase in proportion to the time exposed after the post. Posts that have been registered a long time ago may be distorted as having a large content power because of a long exposed time even if the actual impact is small compared to a recently registered post.

Accordingly, the blog analysis method according to the embodiment of the present invention calculates the user content power by reflecting the influence of the exposure time on the content power of each post. In other words, the content power of each post is corrected according to the exposure time to calculate the user content power. To this end, first, a blog analysis period for calculating user content power is set (for example, 10 days), and the exposure time correction value of the corresponding article is calculated by referring to the time when each post is registered within the analysis period. The exposure time correction value is relatively small when the exposure time is long, and relatively large when the exposure time is short. As an example, the exposure time correction value may be calculated as a ratio obtained by subtracting the exposed time from the analysis time corresponding to the analysis period and dividing the result value by the analysis time. However, the exposure time correction value may be calculated in another way, for example, to be inversely proportional to the exposure time.

The user content power is calculated by multiplying the exposure time correction value calculated for each post by the corresponding post content power and adding the resulting values. Therefore, the content power [UCP (U _i )] of the user U _i is calculated as shown in [Equation 4]. In this case, the value obtained by multiplying the exposure time correction value and the post content power for each post may be referred to as the corrected content power of the post. By calculating the user content power as described above, the distortion of the post content power due to the difference in the exposure time of each post can be reduced, and the accuracy of the user content power can be improved.

Here, ET (D _{i, j} ) is the exposure time correction value of the article D _{i, j} .

Next, an example of calculating user content power will be described with reference to FIG. 6 and [Table 2]. FIG. 6 is a view illustrating exposure time and exposure time correction values of a post to explain an example of calculating user content power according to an embodiment of the present invention.

User U _A registered the posts D _{A, 1} , D _{A, 2} , D _{A, 3} , D _{A, 4} in sequence during the analysis time from the start time of analysis to the end time of analysis, as indicated by the arrows. Power was calculated to be 50, 35, 35 and 10, respectively. The exposure time of each post is the time from the time when the post is registered to the analysis end time, and the exposure time correction values are 0.25, 0.4, 0.8, and 0.95 in order of the longest exposure time. Therefore, the user content power of the user U _A is UCP (U _A ) = 50 × 0.25 + 35 × 0.4 + 35 × 0.8 + 10 × 0.95 = 64.

Referring back to FIG. 4, the user transfer power calculator 137 calculates user transfer power for each user by referring to the propagation history (S460). User U _i of the transmission power, the user U _i can be calculated by adding the product of the weight of the number registered in the own blog and its action through the Reply yeokin to scrap the posts registered in the other user blog, article. If the title La user U _i UDP (U _i) the transmission power of the user, this can be expressed as Equation 5].

UDP (U _i ) = ws2 × SC (U _i ) + wl2 × LC (U _i )

Here, SC (U _i) the user U by _i scrap the posts registered on other people's blogs to their blog many times, LC (U _i) is their respect to the posts registered in the user U _i is other people's blogs The number of times the blog is posted. At this time, ws2 and wl2 are weights for the corresponding actions, respectively, and may be set to be the same as the weights ws1 and ws2 described above.

The propagation probability calculator 138 calculates a propagation probability between users based on the user content power and the user delivery power (S470). The propagation probability is a parameter used in an independent propagation model. According to the propagation probability, a post may be propagated between users.

An independent propagation model and propagation probabilities will be described in detail with reference to FIG. 7. 7 is a schematic diagram illustrating an independent propagation model used in a blog analysis method according to an exemplary embodiment of the present invention.

Blog users can influence each other, and by this influence, blog users can post or register posts registered on other users' blogs to their blogs. For example, if user B reproduces a post owned by user A through scraping or enclosing, user B is said to be assimilated by user A. As such, when a post is spread from one blog to another blog, the probability that the post is spread can be expressed as a probability value, which is called a spread probability. At this time, the probability that user A's contents spread to user B (that is, the probability that user B is assimilated by user A) is how active the user A is in reproducing the content of other users and the content of other users. Determined by

The independent propagation model is a model for explaining the ripple relationship among members in social networks. In other words, the independent propagation model gives a probability to the relationship between blog users, and determines whether to assimilate by this probability when the influence between users. In more detail, the edges connected between blogs in a blog network mean the relationship between blogs, and each edge is given a probability value in the independent propagation model. When a user is assimilated, the neighboring user of the user is assimilated based on the probability value of each edge. In this way, propagation is performed until no user is newly assimilated. The number of users who are assimilated directly or indirectly by the assimilation of a specific user becomes the specific power of the specific user.

For example, referring to FIG. 7, which shows the ripple relationship between blogs in a blog network, when a new post is registered in the blog BA, the post may be spread to the blogs BB and BC with a probability of 0.3 and 0.5, respectively. If a post has been spread to BB, BC, BB and BC can spread the post to another blog that has not yet been spread. In this case, the probability that the blogs BB and BC spread the post to the blog BD is 0.6 and 0.3, respectively. If the blog BD became a fairy tale through this spreading process, three blogs were assimilated by the blog BA, and the ripple power of the blog BA is three. Of course, the ripple probability is randomly determined, so the blogs BB, BC, and BD may not be assimilated, and the ripple power of the blog BA may vary.

In this way, the independent propagation model can quantify the degree to which blog users affect other users, directly or indirectly, through the blog network. In addition, the independent propagation model can adequately describe the ripple relationship between blog users through the propagation of posts in the blog network.

Next, a method for calculating the propagation probability will be described in detail.

The probability that the content of the user U _A to be spread to the user U _B P _{A-> B} is determined on the basis of the user content power UCP (U _A) and the user U _B of the user transmission power UDP (U _B) of the user U _A, It can be calculated as shown in [Equation 6].

P _{A-> B} = wcp × CP (U _A ) + wdp × DP (U _B )

Where CP (U _A ) and DP (U _B ) are normalized values of UCP (U _A ) and UDP (U _B ) from 0 to 1, respectively, and wcp and wdp are normalized user content power and user delivery, respectively. The weight of power. It is possible to normalize the user content power by dividing the content power of the remaining users by the value of the corresponding content power by setting the upper x% or more as 1, and the user delivery power can also be normalized in the same manner as the user content power. It can be based on the upper y%. At this time, the upper x% and y% may be determined in consideration of the characteristics of the blog network to be analyzed. On the other hand, the weights wcp and wdp mean the reflecting ratios of the user content power and the user delivery power, respectively, and satisfy wcp = 1-wdp for convenience of calculation. Therefore, if wcp is determined, wdp is determined to be 1-wcp.

Next, a method of determining the weights wcp and wdp will be described in detail with reference to FIG. 8.

8 is a schematic diagram illustrating an interpolation method for determining user content power weight and user delivery power weight according to an embodiment of the present invention.

The ideal wcp and wdp are the values with the highest accuracy when predicting the ripple by giving the ripple probability between users using the wcp and wdp. In the embodiment of the present invention, Lagrange interpolation is used to determine wcp and wdp having the highest accuracy. That is, wcp and wdp having the highest accuracy can be determined by arbitrarily selecting k wcps and wdps, applying the selected wcps and wdps to predict the ripple, and then predicting the change in the ripple prediction accuracy. The method of calculating the ripple prediction accuracy will be described later.

An example of how to determine wcp and wdp using the Lagrange interpolation method will be described. Referring to FIG. 8, the upper graph shows the distribution of wcp and wdp, where the x-axis represents wcp and the y-axis represents wdp. Since wcp = 1-wdp, (wcp, wdp) has a diagonal distribution. In this example, three values wcp1, wcp2, and wcp3 are randomly selected as wcp to analyze the change in accuracy. The lower graph shows the ripple prediction accuracy obtained through wcp1, wcp2, and wcp3. The Lagrangian interpolation method predicts the ripple prediction accuracy according to the change of wcp. The x-axis represents wcp and the y-axis represents the ripple prediction accuracy according to wcp. As a result, the ripple prediction accuracy has the highest accuracy when wcp is wcp4. Therefore, in this example, wcp4 can be selected as wcp and 1-wcp4 can be selected as wdp.

Of course, you can use other interpolation methods than Lagrangian interpolation. Also, without using interpolation, wcp and wdp having the highest accuracy can be determined by assigning propagation probabilities to all possible wcp and wdp, predicting the propagation, and comparing the accuracy.

Next, a method of calculating the accuracy of the ripple result predicted by the independent propagation model using the wcp and wdp when a ripple probability is provided between users will be described with reference to FIGS. 9 to 11.

9 is a schematic diagram for explaining the response and precision used in the blog analysis method according to an embodiment of the present invention, Figure 10 is an example of the ripple relationship between blog users using a spread history in accordance with an embodiment of the present invention FIG. 11 is a diagram illustrating a blog connection network in which the ripple relationship between users illustrated in FIG. 10 is displayed.

When a particular user writes a post, the accuracy is calculated using the users actually assimilated by the post and the users predicted to assimilate the user through an independent propagation model. Accuracy can be calculated as the average of recall and precision. Responsiveness and precision are concepts used to measure the accuracy of search results in the field of information retrieval. Referring to FIG. 9, the response rate is a ratio of the result set RS, that is, (AS∩RS) / AS, among the answer set (AS), and the precision is the result set (RS). The ratio occupied by the correct answer set (AS), that is, (AS∩RS) / RS. Applying this concept to ripple prediction accuracy, the correct answer set (AS) corresponds to the set of users actually assimilated by the particular user, and the result set (RS) is the set of users predicted to be assimilated using the independent propagation model. Corresponds to.

An example of measuring ripple prediction accuracy will be described with reference to FIGS. 10 and 11. According to the ripple history DH1 and DH3, if user U _A wrote a post, the user who actually spread the post is user U _B , U _C , U _D , U _G , and according to the ripple history DH2 and DH4, user U _C in fact, people who spread the posts case is actually the user who spread the posts by the user _{U B, U D, U E} , U F, according to the ripple history DH5 created that the user U _G posts if written by the user U _C, U _E , U _F. And when user U _A wrote a post, it was predicted that user U _B , U _E would be assimilated through the independent propagation model. When user U _C wrote a post, user U _E , U _F , U _G Let was doerirago moving picture prediction, that the user U _G is through the independent propagation model when writing the posts user U _B, U _C are doerirago moving picture prediction.

Then, the propagation prediction accuracy for the user U _A is that the response is | B | / | (B, C, D, G) | = 1/4, and the precision is | B | / | (B, E) | = 1 Since it is / 2, it becomes (1/4 + 1/2) / 2 = 3/8 (= 0.37). Similarly, the propagation prediction accuracy for the user U _C is that the response is | E, F | / | (B, D, E, F) | = 2/4, and the precision is | E, F | / | (E, F , G) | = 2/3, so that (2/4/4/2/3) / 2 = 7/12 (= 0.58), the ripple prediction accuracy for the user U _G is obtained by the response of | C | / | ( C, E, F) | = 1/3, and the precision is | C | / | (B, C) | = 1/2, so that (1/3 + 1/2) / 2 = 5/12 (= 0.41) do.

By calculating the weighted average by reflecting the number of times the article written by each user actually spreads to the ripple prediction accuracy of each user, the overall ripple prediction accuracy according to the given weights wcp and wdp can be calculated. That is, in the previous example, the number of times the post of the user U _A was actually spread was | DH1 | + | DH3 | = 2 + 3 = 5, and the number of times the post of the user U _C was actually spread was | DH2 | + | DH4 | = 2 + 3 = 5, and the actual number of posts of user U _G was actually spread | DH5 | = 3, so the overall ripple prediction accuracy according to the weights wcp and wdp is (0.37 × 5 + 0.58 × 5 + 0.41 × 3) / (5 + 5 + 3) = 5.98 / 13 = 0.46.

Referring back to FIG. 4, the power user selecting unit 139 applies the propagation probability calculated as in Equation 6 to the independent propagation model and selects the power user based on the propagation probability (S480). In order to find a set of n power users who assimilate the most users in the blog network, the power user selecting unit 139 uses a so-called hill climbing technique. That is, first, the power user selecting unit 139 calculates the propagation power of each user by applying the independent propagation model to the calculated propagation probability. Then, the user who has the highest spread power, that is, the user who is expected to assimilate the most users, is selected, and the user is included in the set of power users. After excluding the users assimilated by the user, the independent propagation model is used again to find the users who are expected to assimilate the most users and include them in the power user set. By repeating this task, we can find a set of n power users. This technique allows the selection of a set of power users within a substantial amount of time while guaranteeing about 63% accuracy.

By selecting the power users, the policy for activating the blog network can be performed. For example, you can focus on promoting your new product to power users, and influence others in your blog network to have a positive awareness of your new product. In addition, it can provide a variety of benefits to power users to encourage them to actively use the blog service to act as a catalyst for the activation of the entire blog network.

In the blog analysis method according to an embodiment of the present invention, the weights ww, wr, wc, ws1 (ws2), wl1 (wl2), wd, and wi may be determined differently depending on which actions are important, for example, 7, 1, 3, 7, 7, 1, 1 can be set respectively. In addition, these weights may be determined in a policy for blog service activation.

Embodiments of the invention include a computer readable medium containing program instructions for performing various computer-implemented operations. This medium records a program or process for executing a method of analyzing a blog as described so far. The media may include, alone or in combination with the program instructions, data files, data structures, and the like. Examples of such media include magnetic media such as hard disks, floppy disks and magnetic tape, optical recording media such as CDs and DVDs, floppy optical and magnetic-optical media, programs such as ROM, RAM and flash memory. Hardware devices configured to store and execute instructions. Alternatively, the medium may be a transmission medium such as an optical or metal wire, a waveguide, or the like including a carrier wave for transmitting a signal specifying a program command, a data structure, and the like. Examples of program instructions include not only machine code generated by a compiler, but also high-level language code that can be executed by a computer using an interpreter or the like.

Although the preferred embodiments of the present invention have been described in detail above, the scope of the present invention is not limited thereto, and various modifications and improvements of those skilled in the art using the basic concepts of the present invention defined in the following claims are also provided. It belongs to the scope of rights.

As described above, according to the present invention, the propagation probability is calculated on the basis of the user content power and the user delivery power of each user in the blog network, and applied to an independent propagation model, thereby establishing various policies for activating the blog service. Power users can be selected.

In addition, according to the spread probability, each blog can be structurally connected in the blog network, and the ripple relationship between blog users can be grasped properly.

Claims (19)

A blog analysis method of a blog analysis system including a blog analysis server, Calculating, by the blog analysis server, user content power of a plurality of first users; Calculating, by the blog analysis server, user delivery power of a plurality of second users; and Calculating, by the blog analysis server, a propagation probability indicating a degree to which a post of the first user spreads to the second user based on the user content power and the user delivery power; Including; The user content power means the degree to which the post of the first user affected the first user and other users in the form of at least one of inquiries, comments, scraps, and loopbacks. The user transmission power refers to the degree to which the second user reproduces at least one of scraping and writing posts of the second user and another user. How to Analyze Your Blog. In claim 1, The user transfer power is a blog analysis calculated by multiplying a scrap weight and a hook weight by multiplying the number of scraps and the number of posts written by the second user with a post of the second user and another user, respectively, and then adding the resulting value. Way. In claim 1, The propagation probability is calculated by multiplying the user content power weight and the user delivery power weight by a value obtained by normalizing the user content power and the user delivery power between 0 and 1, respectively, and adding the resulting value. In claim 3, The user content power is normalized by dividing the value of the user content power corresponding to the top x% by setting the upper x% or more of the user content power to be analyzed as 1 and the user delivery power is the upper y% of the user delivery power to be analyzed. A blog analysis method in which the above is set to 1 and divided by the value of the user delivery power corresponding to the upper y%. In claim 3, And the user content power weight and the user delivered power weight are determined by interpolation to maximize the ripple prediction accuracy according to the ripple probability. In claim 5, The ripple prediction accuracy is calculated as an average of the responsiveness and precision between the user who actually spreads the predetermined article and the user predicted to spread through the independent propagation model. In claim 1, And after the blog analysis server calculates the propagation probability, selecting a set of power users based on the propagation probability and the independent propagation model. In claim 7, The power user set selection step, Calculating the propagation power of the first user by applying the propagation probability to the independent propagation model; Selecting a third user whose maximum power is maximum; and Repeating the ripple power calculation step and the third user selection step a predetermined number except for a user who is expected to be assimilated by the third user. Blog analysis method comprising a. In claim 1, The user content power calculation step, Calculating a post content power of the post based on a blog user's action frequency for at least one post, and Correcting the post content power according to the exposure time of the post Including; The user content power is calculated by adding the corrected article content power. How to Analyze Your Blog. A computer readable medium having recorded thereon a program for causing a computer to execute the method of any one of claims 1 to 9. A user content power calculator configured to calculate user content power of a plurality of first users; A user delivery power calculating unit that calculates user delivery power of the plurality of second users, and A propagation probability calculation unit calculating a propagation probability indicating a degree to which the post of the first user is propagated to the second user based on the user content power and the user delivery power. Including; The user content power means the degree to which the post of the first user affected the first user and other users in the form of at least one of inquiries, comments, scraps, and loopbacks. The user transmission power refers to the degree to which the second user reproduces at least one of scraping and writing posts of the second user and another user. Blog Analytics System. In claim 11, The user transfer power calculating unit multiplies the scrap weight and the chained weight by the number of times the second user scraps and posts the posts of the second user and the other user, respectively, and adds the resulting value to the user transfer. Powered blog analysis system. In claim 11, The ripple probability calculator calculates the ripple probability by multiplying the user content power weight and the user delivered power weight by a value normalized between the user content power and the user delivered power between 0 and 1, respectively, and adding the resultant value. system. In claim 13, The user content power operation unit normalizes the user content power by dividing the user content power by one or more by dividing the user content power by the value of the user content power corresponding to the top x%. And the user transmission power calculating unit sets the upper y% or more of the analysis target user transmission power to 1 and normalizes the user transmission power by dividing the user transmission power by a value of the user transmission power corresponding to the upper y%. In claim 13, And the propagation probability calculator determines the user content power weight and the user transfer power weight by interpolation to maximize the propagation prediction accuracy according to the propagation probability. The method of claim 15, And the spread probability calculator calculates the spread prediction accuracy based on an average of responsiveness and precision between a user who actually spreads a predetermined post and a user predicted to spread through the independent propagation model. In claim 11, And a power user selection unit for selecting a power user set based on the propagation probability and the independent propagation model. The method of claim 17, The power user set selecting unit applies the propagation probability to the independent propagation model to calculate the propagation power of the first user, selects a third user whose maximum propagation power is maximum, and is animated by the third user. Calculating the ripple power and selecting the third user for a predetermined number of users except for a user who is expected to return Blog Analytics System. In claim 11, It further includes a post content power calculation unit for calculating the post content power of the post based on the frequency of the blog user's action on at least one post, The user content power calculation unit corrects the post content power according to the exposure time of the post, and calculates the user content power by adding the corrected post content power. Blog Analytics System.

Images