KR101049648B1 - Blog Rank Method for Efficiently Searching Blogs Using the Blog Rank Algorithm - Google Patents

Abstract

The present invention provides a blog rank method for searching for a blog using a blog rank algorithm, wherein the reputation of the blogger who created the blog page, the trackback connectivity of the blog page, and the user responsiveness of the blog page are evaluated. Evaluates the blog page, but evaluates page e in terms of keyword t, ES (e, t),

ES (e, t) =

Provides a blog rank method for efficiently searching a blog using a blog rank algorithm characterized in that it is defined as.

Description

BLOG RANK METHOD FOR EFFECTIVELY SEARCHING THE BLOG USING BLOG RANK ALGORITHM}

The present invention relates to a blog search method, and more particularly, to a blog rank method for efficiently searching a blog using a blog rank algorithm.

Today, due to the advent of the Web 2.0 environment, most web pages are generated in the blog area, and the existing web pages are also converted into the blog area. Therefore, the importance of efficient information retrieval technology in the blog area is increasing day by day. Among them, page rank algorithm technology is the most important area of such information retrieval technology.

Many algorithms for web page ranking have been studied so far (eg, S. Chakrabarti, 2003. Mining the web, Morgan Kaufmann Publishers.). Among them, PageRank (S. Brin and L. Page, 1998. The Anatomy of a Large-scale Hypertextual Web Search Engine, In Proceedings of 7th International World Wide Web Conference.) And HITS (JM Kleinberg, 1999. Authoritative sources in hyperlinked). environment, Journal of the ACM, Vol. 46, No. 5.) was the most successful result, and the search efficiency for them is well known in the academic and industrial world. The distinctive feature of these algorithms is that they evaluate the superiority of pages based on their connectivity. Of course, these algorithms can be applied directly to the blog area, but it is difficult to expect more efficient search results by overlooking the structural features of the following blog pages.

Unlike web pages, blog pages have structural properties such as trackback connections, tags, and comments (C. Marlow. Audience, structure and authority in the weblog community.In International Communication Association Conference, New Orleans, LA, 2004. http http://web.media.mit.edu/>>ameron/cv/pubs/04-01.pdf.). These attributes are created and modified by bloggers working in the blog area, reflecting their thoughts, interests and responses. In addition, more quality information is generated and managed by inducing interaction with similarly thought and interested bloggers (A. Java, P. Kolari, T. Finin, and T. Oates. Modeling the Spread of Influence) on the Blogosphere.Technical report, University of Maryland, Baltimore County, March 2006.). In other words, the information in the blog area is gradually developed through interaction with these structural attributes. Therefore, these structural attributes and interactions should be considered for more efficient retrieval.

2) A blog site is composed and managed by pages written by one blogger, which is why a blog is often called a “personal publishing tool.” Therefore, the quality and theme of a blog page is usually the knowledge of the blogger who manages the blog. These characteristics make it possible to evaluate a blog that is not initially connected but useful, based on the blogger's past reputation as a good page.

In other words, most web pages are generated in the blog area and the existing web pages are being converted into the blog area. In this situation, the characteristic differences between blog pages and web pages, such as trackback connections, blogger reputation, tags, and user responsiveness, are an important issue that cannot be overlooked. Also, simply applying a traditional web page ranking algorithm to a blog page that does not reflect this characteristic difference is not appropriate for efficient search.

The present invention has been invented in view of the above, and utilizes the structural features of the blog to evaluate the blog's reputation, trackback connectivity, and user responsiveness, and based on the blog rank algorithm that efficiently evaluates the blog page. The purpose is to provide a blog ranking method for searching.

A blog rank method for efficiently searching a blog by using a blog rank algorithm according to the present invention for achieving the above object,

In a blog rank method for searching a blog using a blog rank algorithm,

Evaluate the blog page's reputation, the trackback connectivity of the blog page, and the user responsiveness of the blog page.

In terms of keyword t, the evaluation of page e, ES (e, t),

ES (e, t) =

here,

b: blogger who created page e with tag t

K: number of trackback connections on page e

b _i : Blogger creating the i th trackback connection on page e

BS (t, b): Blogger b's reputation score for tag t

TBS (t, e, b _i ): In terms of tag t, generated by blogger b _i contained in page e

               Score of the i trackback connection

로서,

as,

N _tr (b, t): for all pages written by blogger b for tag t

          Trackback Connections

N _cm (b, t): Number of comments on all pages written by blogger b about tag t

β: 0.001 (for no trackback connection)

CS (e): Number of comments on page e

It is characterized by.

According to the present invention, through comprehensive analysis of the blog structure, implicitly reflects not only physical connectivity but also content evaluation through user participation, thereby improving search efficiency.

In addition, it will be more efficient to evaluate pages that do not have links at the beginning of their writing.

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

The blog rank method according to the present invention uses a "blog-rank" algorithm that evaluates a page in consideration of characteristics of a blog. This algorithm evaluates pages with higher trackback connectivity and user responsiveness as better pages, and also pages with better reputation bloggers. Pages have more trackback connectivity, with more trackback connections from bloggers with a higher reputation. In order to realize this blog-rank algorithm, the following three factors are evaluated in the present invention.

The reputation of the blogger who wrote the page

Page Trackback Connectivity

User responsiveness of the page

The blogger's reputation is evaluated by calculating two scores based on trackback connections, tags, and comments. The score of pages written by bloggers, and the score of blogger activity in the blog area. Page scores are calculated based on tags, trackback links, and comments, and blogger activity scores are based on the number of good pages written by bloggers. Good pages in the present invention are mainly pages with high trackback connectivity.

The trackback connectivity described above is calculated based on the number of trackback connections and the reputation of the blogger that created the connection.

The user responsiveness described above is calculated based on the valuation of the comment on the trackback connection.

The blog-rank algorithm according to the present invention has two important features.

The first feature is that when evaluating a page, it evaluates from a structural point of view. In other words, we are not interested in the contents of the page. However, because the structural attributes of the blog reflect the quality of the page, the content evaluation is implicit. For example, if a blogger sees a useful and interesting page on a blog site, he or she would like to make that page available to users of his site, thereby creating a trackback link with his site. Or, even if you don't share via trackback link, you will express your opinion on page through comments. In other words, trackback links and comments indicate that the content of the page is useful and interesting to someone. After all, a page that contains more of these attributes or that contains better ones is likely to be more informative for users. In other words, the content evaluation of blog pages is implicitly performed by the activities of bloggers, and the evaluation results are naturally included in the structure of blog pages.

Another feature of the blog-rank algorithm according to the present invention is that the page can be evaluated based on past blogger activity. This can be highly appreciated for bloggers who don't have internal links for the first time, but are likely to be useful information. This feature is very important considering the characteristics of blog areas that mainly deal with new topics.

So far there has been some research for blog page ranking. For example, Adar et al. Proposed a ranking method called iRank [E. Adar., L. Zhang., L. Adamic., R. Lukose., (2004). "Implicit Structure and the Dynamics of Blogspace", Workshop on the Weblogging Ecosystem: Aggregation, Analysis and Dynamics, WWW 2004].

While this method highly evaluates a page including a source of information, the ranking method using the blog-rank algorithm according to the present invention is a method of highly evaluating a famous page.

The ranking method using the blog-rank algorithm according to the present invention, which ranks famous pages higher based on the user's interest and interaction, provides better information due to the characteristics of the blog area that mainly deal with the latest topics.

Meanwhile, according to Fujimura et al. [K. Fujimura., T. Inoue., M. Sugisaki., (2005). "TheEigenRumor Algorithm for Ranking Weblogs", 2nd Annual Workshop on the Weblogging Ecosystem: Aggregation, Analysis and Dynamics, WWW 2005.]. We are suggesting how to evaluate. This method does not consider trackback connectivity as it relies solely on the provision and evaluation of information. On the other hand, since the connectivity evaluation of the blog-rank algorithm according to the present invention is based on the trackback connectivity, implicit inclusion of the content evaluation of the page enables better information retrieval.

Meanwhile, the connectivity analysis of the blog-rank algorithm according to the present invention is similar to PageRank and HITS. However, since the blog-rank algorithm according to the present invention evaluates and manages the blogger's reputation score, the page may be evaluated according to the blogger's reputation evaluation regardless of when the page is created. This is almost impossible with PageRank, HITS, which requires a certain amount of time to score points for evaluation.

In addition, by implementing a software search module applying the blog-rank algorithm according to the present invention compared the search method according to the existing algorithm and the search performance. For this comparison, we collected 1,95 blog sites and 62,906 pages from the Tistory domain (http://www.tistory.com/). As a result of the search comparison, the search method applying the blog-rank algorithm according to the present invention searched more relevant information for the user query than the method based on the existing Tstory search system. This shows that the structural features of blogs can improve the performance and usefulness of blog search.

1 illustrates a structural feature of a blog area, where a blog page is structurally different from an existing web page. In other words, the blog area has a large number of blog sites. A blog site consists of a blogger and the pages he writes. A page is composed of attributes such as content, tag, trackback link, comment, ID of the blogger who created the page, and creation time. In addition, bloggers in the blog area read two pages in the blog area and create trackback links and comments to create two kinds of interactions between blog sites. Trackback links are links between blog pages and other blog sites, and comments are interactions between blog pages and users.

Also, users usually use keywords when searching on the web, and the search engine provides the user with information related to the keywords. In general, blog pages contain tags (C. Marlow. Audience, structure and authority in the weblog community.In International Communication Association Conference, New Orleans, LA, 2004. http://web.media.mit.edu/ >> ameron / cv / pubs / 04-01.pdf.). Also, bloggers usually choose tags carefully to better represent their page content. Therefore, the present invention uses a tag for evaluating the correlation between the user's query keyword and the content of the page. That is, if keyword k is present as a tag on a page, the content of that page is assumed to be related to the query keyword.

Finally, in the present invention, a good page is determined according to the following assumptions.

Pages written by better bloggers are better

The more trackbacks generated by better bloggers, the better pages

Better pages with more comments

Taking these factors into consideration, the page score is evaluated as in Equation (1) below.

In other words,

Score on page with tag t = blogger's reputation score for keyword t +

                              Trackback's Score + Comment's Score ------- (1)

Here, the blogger's reputation score for the keyword t is evaluated based on the page written by the blogger including the keyword t and the past activity of the blogger for the keyword t.

Next, a blog-rank algorithm according to the present invention for evaluating each element will be described.

1. Blog-rank algorithm

According to the above evaluation formula (1), the blog-rank algorithm is defined as the following formula (2).

In terms of keyword t, evaluation of page e, ES (e, t),

-------- (2)ES (e, t) =

-------- (2)

here,

b: blogger who created page e with tag t

K: number of trackback connections on page e

b _i : Blogger creating the i th trackback connection on page e

BS (t, b): Blogger b's reputation score for tag t

TBS (t, e, b _i ): In terms of tag t, generated by blogger b _i contained in page e

               Score of the i trackback connection

로서,

as,

N _tr (b, t): for all pages written by blogger b for tag t

           Trackback Connections

N _cm (b, t): Number of comments on all pages written by blogger b about tag t

β: 0.001 (for no trackback connection)

CS (e): Number of comments on page e.

α ^b _t is the weight of the comment. Since the value of the comment is smaller than the value of the trackback, α ^b _t weakens the influence of the comment. Equation (2) indicates that the blog-rank algorithm evaluates a page based on characteristics of the blog structure such as bloggers, trackback links, comments, and tags for efficient blog search. Hereinafter, each of the above elements will be described.

1.1 Blogger Reputation Score (BS)

The blogger reputation score indicates the blogger's reputation for a particular tag. The method according to the invention assumes that the blogger's reputation for a subject is based on his past behavior and past output. If a blogger is active in the blog area on a topic (tag), and the output of the activity is good, his reputation will be high.

To do this, we consider two points of view: the average score (BES) of all pages written by the blogger for a particular tag, and the blogger activity score in the blog area for that particular tag. Normalize both scores to prevent one score from significantly affecting the blogger reputation score. Based on these points, the blogger reputation score is defined as follows.

BS (t, b) = sigmoid (BES (t, b)) + sigmoid (BAS (t, b))

Where sigmoid (a) = 1/1 + e ^-a

1.1.1 Page Scores for Bloggers

The blogger's page score indicates the quality of the blogger's pages for a particular tag. In the present invention, the page score of the blogger is defined as the average score of the pages written by the blogger in the past. This is because pages written in the blog area are the main output.

Where E ^b _t : the set of all pages written by blogger b for tag t

In the present invention, a page written in the past by a blogger is assumed to be a page written without a blogger's reputation. Therefore, in the score calculation formula (2) of the page, the blogger reputation score (BS) is set to the default value of 1, and the score (PES) of the previously created page is defined as follows.

here,

K: Number of all trackback connections on page e

b _j : Blogger creating j trackback link on page e _i

BS (t, b): Blogger b's reputation score for tag t

TBS _j (t, e _i , b _j ): In terms of tag t, generated by blogger b _j contained in page e _i

                Score of the i trackback connection

로서,

as,

N _tr (b, t): for all pages written by blogger b for tag t

          Trackback Connections

N _cm (b, t): Number of comments on all pages written by blogger b about tag t

β: 0.001 (for no trackback connection)

CS (e): Number of comments on page e.

Since the weights α and β are very small values, CS (e _i ) does not significantly affect the calculation of the pages created in the past. Therefore, the page score of the blogger is greatly influenced by the trackback score and the blogger activity score. Here, the blogger activity score is also based on the number of trackback connections, and consequently the overall evaluation of the blog-rank algorithm is most affected by the trackback connections. The trackback score will be described later.

1.1.2 Blogger Activity Score (BAS)

The blogger activity score represents blog activity in the blog area for a particular tag. This blogger activity score is calculated based on the number of all pages written by the blogger and the trackback value of one page per full page written by the blogger. To realize this idea, the Blogger Activity Score (BAS) is defined as follows.

----- (4)

----- (4)

here,

N ^t _en (b): Number of all pages written by blogger b for tag t

N ^tr _en (b): Total number of pages written by blogger b that have trackback connections

N _en (b): Number of all pages written by blogger b.

In Equation (4), the first element represents the amount of blogger behavior and the last element represents the quality of blogger behavior. The quality of blogger behavior is based on trackback connections.

1.2 Trackback Score (TBS)

The trackback score indicates the trackback connectivity of the page with a specific tag. As mentioned above, trackback connectivity is evaluated based on the number of trackback connections and the reputation of the blogger that created the trackback connection. Thus, in terms of tag t, the trackback score of page e _i is defined as follows:

TBS _i (t, e _i , b _j ) =

here,

b _j : blogger that created the jth trackback connection b

BS (t, b _j ): The reputation score of blogger b that created the jth trackback connection.

The value 1 represents the quantitative component of the trackback, and BS (t, b _j ) represents the qualitative component of the trackback.

Experiment setup

The ultimate goal of the blog-rank algorithm is to improve the association of query keywords with the returned search results. To demonstrate this improvement, 195 blogs and 62906 pages were collected from the Tstory domain to implement a blog search method using the blog-ranking algorithm according to the present invention. In addition, the blog search results using the blog-ranking algorithm according to the present invention were compared with the existing blog search results.

For more efficient search results, pages that are more relevant to the query keyword should be ranked higher than those that are not. As mentioned above, bloggers carefully choose tags to better represent their page content. Therefore, in the present invention, it is assumed that a page including a tag related to a query keyword is related to the query keyword. In other words, if a page contains more tags that are related to a query keyword, the page is more relevant to the query keyword.

Tags associated with query keywords are defined based on the co-occurrence frequency. If there is a page that contains the same tag as the query keyword, all tags in that page have an association of 1 with the query keyword. As such, if the tags having an association of 1 are present on another page at the same time, the association value is increased by one. This process is repeated to select tags with relevance greater than 10 and define them as tags related to the query keyword. In the present invention, these tags are assigned to the Tstory domain (http://www.tistory.com/), the Eagles domain (http://www.egloos.com/), and the Blog Korea domain (http://www.blogkorea.net/ We configure in page of). Here, the story is a domain that has been collected for computing the blog-rank ranking algorithm according to the present invention, and Eagles and Blog Korea are domains that are not.

To evaluate the effectiveness of the search results, we measured the number of tags associated with the query keywords included in the page and applied the number to NDCG at K (Normalized Discounted Cumulative Gain). NDCG at K is one of the evaluation metric of IR which measures rank accuracy of search results [K. Jarvelin and J. Kekalainen. IR evaluation methods for retrieving highly relevant documents. In Proceedings of the ACM Conference on Research and Development on Information Retrieval (SIGIR), 2000]. When there is a given query q, the ranked search result is NDCG calculated by K rank by the following equation.

NDCG at K is calculated as the sum of gains from rank j = 1 to K in the search results. r (j) is a function representing the reward in rank j. In an experiment of the blog-ranking method according to the present invention, r (j) is calculated based on the RV (e _j, q) of the j th page e _j . That is, r (j) = log (RV (e _j, q) + 1). RV (e _j, q) is the number of tags associated with the query keyword q in page e _j . M _q is a normalization constant, which causes the maximum value of NDCG to be 1.

For the practicality of the experiment, a query keyword randomly selected from the query keyword pool was used. A total of 20 experimental queries consist of 10 issue queries and 10 random queries. The issue query has recently selected keywords related to events or characters that are social issues in the query keyword pool, such as Lee Myung-bak, mad cow disease, and Olympics. Random queries were randomly selected from the query keyword pool. The query keyword pool consists of the following steps.

(1) Set the set with the initial keyword as a query-set. Initial keywords are randomly selected.

(2) Randomly select a keyword from the query-set.

(3) Search the blog area for keywords related to the selected keyword, and add the found keywords to the query-set. Related keywords are all tags on the page that have the same tag as the selected keyword.

(4) Abort if the size of the query-set exceeds 100.

(5) Select one keyword not previously selected and go back to step 3.

Use the completed query-set as a query keyword pool.

Experiment result

In order to evaluate the search efficiency of the blog search method according to the blog-rank algorithm according to the present invention, the following experiment was set as follows.

(1) Select experimental queries. The experimental query consists of 10 issue queries and 10 random queries.

(2) The experimental queries are applied to the blog-rank search method and the Tstory blog search method according to the present invention, and receive a search result.

(3) The NDCG at K evaluation metric and related tags in 3 domains are used to compare and analyze the results of the two methods.

First, we experimented with the related tags configured in the story domain.

2A and 2B are graphs showing search performance improvement of each of an issue query and an arbitrary query based on the related tags of the present invention and the tistory domain.

As shown in FIG. 2A, it can be seen that continuous improvement of overall performance is shown in the top ten pages. In FIG. 2A, the NDCG value represents an average value of ten issue queries. In rank 2 (K = 2), it can be seen that the NDCG value of the blog-rank according to the present invention is 0.804, and the value of the tistory-rank is 0.615.

In addition, as shown in FIG. 2B, it can be seen that a continuous improvement in overall performance occurs in the top 10 page even in arbitrary queries. The biggest performance improvement was found in rank 5, while the blog-rank according to the present invention was 0.764, while the story-rank was 0.645.

These results show that the blog-rank algorithm according to the present invention has better search performance than the conventional rank algorithm. In addition, performance improvements in issue queries are better than performance improvements in arbitrary queries, showing that the blog-rank algorithm better reflects the recently-issued pages.

In addition, experiments were performed by applying related tags configured in two domains other than the experimental domain of the present invention, that is, the igloos domain and the blog Korea domain.

3A and 3B are graphs showing the improvement of search performance of an issue query and an arbitrary query based on the related tags of the present invention and the eagles domain, and FIGS. 4A and 4B are based on the related tags of the present invention and the blog Korea domain. This is a graph showing the improvement of search performance for each issue query and random query.

In experiments using tags from Eagles and Blog Korea, it can be seen that the performance of the blog-rank algorithm according to the present invention is better.

As a result, it was proved that the search system to which the blog-rank algorithm according to the present invention is applied is superior to the search performance of the existing Tstory search system.

Table 1 shows the performance improvement of the blog-rank algorithm according to the present invention. In Table 1, each value is a difference between the NDCG value of the method of applying the blog-rank algorithm and the search method according to the tistory, and the values in parentheses are performance improvement rates.

                            TABLE 1

 Configuration domain of associated tags Tistory Eagles BlogKorea Issue query  0.168 (24%)  0.129 (18%)  0.142 (20%) Random query  0.085 (12%)  0.057 (7%)  0.090 (15%)

As described above, Table 1 shows the experimental results to which the entire test query is applied, and the blog rank method using the blog-rank algorithm according to the present invention shows that the entire query (issue query + random query) is better than the conventional search method. The average 16% performance improvement was achieved, and the performance improvement on the issue query was 20% on average, which is 2 times higher than the average improvement of 11% on arbitrary queries.

As mentioned above, unlike traditional web pages, blog pages have structural features such as trackback links, tags, and comments. Therefore, the blog rank algorithm according to the present invention that considers these features has better blog search performance than the existing web page rank algorithm.

Moreover, the algorithm according to the present invention analyzed the blog page into three main factors: blogger's reputation, trackback connectivity, and user's responsiveness. Among these factors, trackback connectivity and user responsiveness only evaluate the structural elements of the blog, but provide better search performance, including implicit content evaluation. This is one of the important features in terms of computing performance. Because analyzing the content of data directly requires expensive computing costs. In addition, the method according to the present invention evaluates and manages the blogger's reputation so as not to underestimate pages that are not initially connected but have useful content.

1 is a view showing the structural features of the blog area;

2A and 2B are graphs illustrating an improvement in search performance of each of an issue query and an arbitrary query based on the related tags of the present invention and the tistory domain,

3A and 3B are graphs illustrating an improvement in search performance of each of an issue query and an arbitrary query based on related tags of the present invention and the eagles domain,

4A and 4B are graphs illustrating search performance improvement of each of an issue query and an arbitrary query based on related tags of the present invention and the blog Korea domain.

Claims (6)

In a blog rank method for searching a blog using a blog rank algorithm, Evaluate the blog page's reputation, the trackback connectivity of the blog page, and the user responsiveness of the blog page. In terms of keyword t, the evaluation of page e, ES (e, t), ES (e, t) =

here, b: blogger who created page e with tag t K: number of trackback connections on page e b _i : Blogger creating the i th trackback connection on page e BS (t, b): Blogger b's reputation score for tag t TBS (t, e, b _i ): In terms of tag t, generated by blogger b _i contained in page e                Score of the i trackback connection

as, N _tr (b, t): for all pages written by blogger b for tag t           Trackback Connections N _cm (b, t): Number of comments on all pages written by blogger b about tag t β: 0.001 (for no trackback connection) CS (e): Number of comments on page e Blog rank method for efficiently searching a blog using a blog rank algorithm, characterized in that defined as. The reputation score BS (t, b) of blogger b for the tag t, BS (t, b) = sigmoid (BES (t, b)) + sigmoid (BAS (t, b)) Where sigmoid (a) = 1/1 + e ^-a Blog rank method for efficiently searching a blog using a blog rank algorithm, characterized in that defined as. 3. The blogger's page score BES, according to claim 2,

Where E ^b _t : the set of all pages written by blogger b for tag t Blog rank method for efficiently searching a blog using a blog rank algorithm, characterized in that defined as. The method of claim 1, The formula ES (e, t) =

Sets blogger reputation score (BS) to the default value of 1 to create a score PES for the past page, and the score PES for the past page,

here, K: Number of all trackback connections on page e b _j : Blogger creating j trackback link on page e _i TBS _j (t, e _i , b _j ): In terms of tag t, the blogger b _j contained in page e _i is generated.                 Score of successful i-th trackback connection

as, N _tr (b, t): Trackback of all pages written by blogger b for tag t            Number of connections N _cm (b, t): Number of comments on all pages written by blogger b about tag t β: 0.001 (for no trackback connection) CS (e): Number of comments on page e Blog rank method for efficiently searching a blog using a blog rank algorithm, characterized in that defined as. The method of claim 2, wherein the blogger activity score BAS is

here, N ^t _en (b): Number of all pages written by blogger b for tag t N ^tr _en (b): Total number of pages written by blogger b that have trackback connections N _en (b): number of all pages written by blogger b Blog rank method for efficiently searching a blog using a blog rank algorithm, characterized in that defined as. The system of claim 1, wherein the trackback score TBS is: TBS _i (t, e _i , b _j ) =

here, b _j : blogger that created the jth trackback connection b BS (t, b _j ): The reputation score of blogger b that created the jth trackback connection. Blog rank method for efficiently searching a blog using a blog rank algorithm, characterized in that defined as.

Images