WO2008023904A1

WO2008023904A1 - Document ranking granting method and computer readable record medium thereof

Info

Publication number: WO2008023904A1
Application number: PCT/KR2007/003956
Authority: WO
Inventors: Seok Hoo Hong; Kyung Seok Jeong; Sang Ho Lee; Kil Jae Lee; Seong Min Jang; Young Sin Jung; Hyun Ae Yoo
Original assignee: Sk Communications Co., Ltd.
Priority date: 2006-08-22
Filing date: 2007-08-20
Publication date: 2008-02-28
Also published as: KR20080017685A; KR100818553B1

Abstract

The present invention relates to a document rank granting method, and more particularly to a method for granting document rankings in such a manner that when a user inputs a search query through a search window of a search engine, a document for which the number of actions by other users interested in the query is larger and the user of which is a guru in the subject is ranked at a higher level, by reflecting both an action ranking of the user of a document corresponding to the query and a document composition time in the weight of the document. The present invention shows documents and users that coincide with a specific query, and provides far superior search performance as compared with the existing simple keyword-based search ranking method, by using feedback information from users.

Description

DOCUMENT RANKING GRANTING METHOD AND COMPUTER READABLE

RECORD MEDIUM THEREOF

Technical Field The present invention relates to a document rank granting method, and more particularly to a method for granting document rankings, which shows documents and users that coincide with a specific query, and provides far superior search performance as compared with the existing simple keyword-based search ranking method, by using feedback information from users, in such a manner that when a user inputs a search query through a search window of a search engine, a document for which the number of actions by other users interested in the query is larger and the user of which is a guru in the subject is ranked at a higher level by reflecting both an action ranking of the user of a document corresponding to the query and a document composition time in the weight of the document.

Background Art

When users search for information by means of search engines, the most important one of various quality measures of the search engines is "ranking." Here, the term "ranking" represents granting ranking to suitable documents based on specific keywords, thereby locating suitable documents at upper positions so as to reduce the user effort for information acquisition.

Currently, the Korean portal sites use either appearance location information or the frequency of appearance in search result documents based on a searched query. However, by only employing this scheme, it is difficult to perform an efficient search enabling the user to easily find one desired document from among a huge number of documents .

Other countries' portal sites employ various weight determination schemes using multiple pieces of meta-information and so on in HTML, starting with the page ranks using the inter-document link structure, in addition to word-based appearance information. Also, the meta-search engines may employ a method of selecting a search system according to queries, and so on. However, such a method provides only very limited performance improvement at a high cost, and cannot be collectively applied to various contents provided by the Korean portal sites because it has been verified in terms of web pages .

In documents and theses published until recently, studies of assigning appropriate weights with respect to multiple elements taking search rankings into consideration have been conducted. As a result, it has been verified that the page rank exerts the best effect based on web pages, and the most powerful element varies according to domains.

However, these methods are not suitable for searching User-Created Contents (UCC) types of documents, which are among the principal documents currently provided in Korea, and require a high cost, so that there is difficulty in applying the methods.

Technical Solution

The present invention has been made to solve the above-mentioned problems occurring in the prior art, and the principal factors for ranking contents of Korean portals include subject suitability and user's action information. In order to provide documents having a subject suitable to a specific query, and information about gurus suitable thereto, an automatic classification method is required. According to the automatic classification method, a predetermined subject list is defined, documents are analyzed, and the documents are matched with subjects that can represent the documents. In addition, an object of the present invention is to provide a method for calculating a ranking value in connection with previous subject classification suitability so as to place user's favorite documents at the upper positions, by means of the action information of the user.

The present invention proposes "ranking"

(hereinafter, referred to as "action ranking") which can place documents closely coinciding with a subject or user' s favorite documents at upper positions by using documents which accord with a keyword input by the user or have subjects similar to the subject, together with various user feedback information (hereinafter, referred to as an "action") . In addition, the present invention is to disclose a list of gurus (professionals) suitable to an input keyword, thereby maximizing the efficiency of search.

According to an aspect of the present invention, there is provided a method for granting a document ranking, the method including: a document composition step of composing a document by a user; a query input step of inputting a query by a searcher; a document weight extraction step of extracting a document weight, in consideration of both a probability that the document composed in the document composition step coincides with the query, and a sum of a plus influencing power, a duplicate influencing power, and a click influencing power up to a search date for the document; a user ranking extraction step of extracting a user ranking, from among a total sum of document weights which are caused by other persons' actions with respect to documents connected with the query among documents composed by the user, a total sum of plus weights by the user with respect to other persons connected with the query, a total sum of duplicate weights, and a total sum of click weights; and a document ranking extraction step of extracting a document ranking from the document weight and the user ranking.

Description of Drawings

FIG. 1 is a block diagram illustrating the configuration of a ranking calculation method applied to the present invention;

FIG. 2 illustrates a table showing types of actions applied to the present invention, and information about whether each action is used or not;

FIG. 3 is a graph illustrating a time weight as a function of a time period "d_c-d_w" passed from a date when document is generated; FIG. 4 is a view of an application method upon a search which is used in the present invention;

FIG. 5 is a view illustrating a search result screen to which the present invention is applied; and

FIG. 6 is a view illustrating an expert result screen to which the present invention is applied.

Mode for Invention

Preferably, the document ranking is calculated by an equation,

wherein the ^XΛW(s,d_j)" represents a document weight, and the "AR_A(s,ui)" represents a user ranking.

Preferably, the document weight "W(s,d_j)" is calculated by an equation, W(s,d_j)=P(s,d_j)x(aχP(d)+βχR(d)+rxC(d))_r

wherein the "α", "β", and ^Λλγ" represent weights for respective actions; the ^NλP(s,d_j)" represents a probability that a j^th document coincides with a subject "s"; the "P(d)" represents a total of plus influencing powers up to date "d" and is equal to "P' (d-1) +p (d) , " in which the ^XλP' (d-1)" represents a total of plus influencing powers up to one day before date "d, " and the "p(d)" represents a plus influencing power of another person on date "d"; the "R(d)" represents a total of duplicate influencing powers up to date "d" and is equal to "R' (d-1) +r (d) , " in which the "R' (d-1)" represents a total of duplicate influencing powers up to one day before date "d, " and the ^λλr(d)" represents a duplicate influencing power of another person on date "d"; and the ^λλC(d)" represents a total of click influencing powers up to date ^λxd" and is equal to "C (d- l)+c(d)," in which the "C (d-1)" represents a total of click influencing powers up to one day before date "d, " and the "c(d)" represents a click influencing power of another person on date "d."

Preferably, the user weight "AR_A(s,ui)" is calculated by an equation,

wherein the document weight is defined by W(s,d_j)=P(s,d_j)x(axP(d)+βxR(d)+γxC(d)) ; the plus weight is

defined by W{s,p_i)=P{s,d_l)xAR{d_l,u_i) ; the duplicate weight is

defined by W(s,r_o)=P(s,d_o)xAR(d₀,U₁); and the click weight is

defined by W(s,c_m)=P(s,d_m)xΛΛ(d_m,M,) , wherein the "P (s, Cl₁)" represents a probability that an 1^th document coincides with a subject "s"; the "AR(di,ui)" represents an action ranking score of an owner of a document ^Xλd, " with respect to the subject "s," to which the document ^ΛΛd" belongs; the ^NNP(s,d_o)" represents a probability that an o^th document coincides with a subject "s"; the ^λλAR(d_o,Ui)" represents an action ranking score of an owner of a document "d," with respect to the subject "s," to which the document "d" belongs; the "P(s,d_m)" represents a probability that an m^th document coincides with a subject ^λλs"; and the "AR(d_m,Ui)" represents an action ranking score of an owner of a document "d," with respect to the subject "s," to which the document "d" belongs. Preferably, a time weight is applied to at least one of the user ranking and document ranking so that the corresponding ranking becomes lower as time passes.

Preferably, the time weight is defined by a formula, and the application means that the user ranking and document ranking are multiplied by the time weight,

0.998^Λ (d_c-d_w)_f

wherein the "d_c" represents a time when an action ranking is calculated, and the "d_w" represents a time when a document is created. Preferably, a final document ranking is calculated by applying a system ranking to the document ranking.

Preferably, the final document ranking "

"

is calculated by an equation,

wherein the

represents a final document ranking for a query "q"; the "SJ" represents a j^th subject value; the "dj_." represents an i^th document; the

Wq„1"1 represents a query; the "ave" represents an average

value; the «

TM represents a cosine value between a

subject vector "S_j" and a query vector "q"; the ^Xλ AR_A(s_j,d_{) " represents an A-type action rank score of an

i^th document of documents belonging to a j^th subject; the "α" and "β" represent an action ranking score and a SR' s weight; and the "SR" represents a system rank value (i.e., a search engine's ranking score).

Preferably, the method further includes a document ranking extraction step of extracting a document ranking for a document of which an author is unknown, in which the document ranking for the document of which the author is unknown is extracted in consideration of influencing powers of a number of times of pulses and a number of times of clicks with respect to the document, and a system rank value.

Preferably, the document ranking ^XλAR_B(s,dj)" for the document of which the author is unknown is calculated by an equation,

AR_B(s,d_t) =axP"+βxC"+χxSR+k , wherein the P" represents a number of times of pluses, the C" represents a number of times of clicks, the SR represents a system rank value, and the k represents a constant.

Preferably, a time weight is applied to the document ranking of the document of which the author is unknown, so that the ranking becomes lower as time passes.

Preferably, the time weight is defined by a formula, and the application means that the document ranking of the document, of which the author is unknown, is multiplied by the time weight,

0.998^Λ (d_c-d_w), wherein the "d_c" represents a time when an action ranking is calculated, and the "d_w" represents a time when a document is created.

In accordance with another aspect of the present invention, there is provided a method for granting a document ranking, the method including the steps of: calculating a weight for a first document according to a user's action applied to the first document; calculating a first degree of matching between the calculated weight for the first document and a first keyword; and arranging the first document and a second document according to the first degree of matching and a second degree of matching when a query by the first keyword is generated, the second document having the second degree of matching equal to or greater than a predetermined value with respect to the first keyword.

Preferably, the weight for the first document is proportional to a ranking of a user who composes the first document.

Preferably, the weight for the first document is proportional to a number of times of bookmarks of the first document or a number of times of replies added to the first document, by a user reading the first document .

Preferably, the weight for the first document is proportional to a number of users who have read the first document.

Preferably, the present invention further includes a step of setting the degree of matching of the first document with respect to the first keyword to a higher value than the degree of matching of the second document, when the first and second documents include the same contents, and an original author of the same contents is an author of the first document. Preferably, the weight for the first document is proportional to a temporal sequence of a user' s action generated with respect to the first document.

In accordance with still another aspect of the present invention, there is provided a computer-readable recording medium in which a program for implementing the aforementioned method is recorded.

Hereinafter, the preferred embodiments of the present invention will be described in detain with reference to the accompanying drawings. However, these drawings are for illustrative purposes only, and the present invention is not limited thereto.

FIG. 1 shows a database for storing information about documents to be searched for, as well as user feedback information, an automatic classifier for performing automatic classification, and a ranking calculation module for calculating a ranking value by means of classification information and user action information. An "action" implies both a suggestive activity and an explicit activity in the course of information search. Such an action may be classified into two types: the first is a creation action as an information provision subject, and the second is a delivery action as an information searcher. Although the present invention will deal importantly with both types, a higher value is granted to the creation action, rather than the delivery action in terms of actual action weights . FIG. 2 illustrates a table showing types of actions and adopted actions.

A write action represents an action of writing a document. The write action implies a piece of writing composed on a Tong, a Paper, or a Cyworld board by the user, and is the most important of all actions. Upon subject classification, a write action value, which is a value to classify a document into a specific subject, is re-adjusted. "Wiki" refers to a service to be opened like a search plus service, in which a weight may be granted to a piece of writing composed by the user. Since users capable of composing a piece of writing in "Wiki" are limitative, it is unreasonable to give the same weight to "Wiki" as it does to the write action. This violates the rule that all users must compete with each other under the same conditions. Gurus become higher-ranking gurus, and the probability of the general users entering a guru group is reduced, so that the phenomenon of the rich getting richer and the poor getting poorer may be caused, even in this case. Also, in "Wiki," since one document may be composed by multiple users, it is not easy to reflect the action rankings. Therefore, "Wiki" is limited to only being used as an encyclopedia, and is handled without regard to the action ranking.

An RSS action refers to a score which is absorbed into my influencing power from the influence index of another user who periodically reads my archives through the RSS. However, although the RSS action may reflect only the case of registering documents of specific subjects, the whole registered through the RSS cannot be mapped to a specific subject, so the RSS action is excluded from the action ranking, for the moment.

A plus action is a concept of a bookmark for a good document found as a result of a search or during Web surfing, in which a user's plus action for a document is reflected in an action score of the owner of the document, and also, the contents of the document are slightly reflected in my action score. The plus action includes estimation for the document, and exerts influence, even on the ranking value, according to a degree of quality.

A comment action refers to reflecting the number of other persons' comments for a specific document on the action ranking of the owner of the specific document. Since there may be no comment system, depending on the kinds of services where documents are created, it is unreasonable to identically apply a comment value to all documents. Also, since it is difficult for a system to determine whether the contents of each comment are affirmative or negative, it is risky and uneconomic for the cost to directly reflect the comment action in the ranking. A query action refers to a weight for a keyword used as a search query by a user. Although the query action helps to identify the propensities of users - which user is interested in which subject - a large number of queries do not always mean high information power, so that the query action is not available in terms of information power.

A click action is related to documents of other persons which are included in a search result list and are clicked by a user. In this case, my influencing power is reflected in a clicked document, and the contents of the document are slightly reflected in my propensity and action score.

A duplicate action is applicable only to A-type documents (Netizen Information Best). The duplicate action is intended to give a relatively higher score to the original author by picking out a scrap, a surreptitious use of a partial paragraph, etc., and to supply a document having a higher duplication rate with a relatively higher write action value. This can be reflected by the plus action, upon an actual operation. For example, when user ^λλA" pastes a document of user "B" into a document of the user "A, " it can be calculated as user A' s plus action with document "B." An action ranking calculation method requires automatically classifying user documents (or actions) according to subjects in order to calculate the rankings of the documents according to the subjects. Thereafter, in the classified documents, the influencing power of another person performing an action is reflected. The calculation procedure is as follows:

- Automatically classify an action

- Reflect an influencing power of another person on a probability value of the classified action - Take scores of actions in which influencing powers are reflected

An action is based on a document and a keyword. Since, even in a plus or click action, the subject thereof is a document, the automatic classification is performed with documents and keywords as basic sources. With respect to automatically classified actions, degrees of matching with one subject are expressed as probability values. In addition to the probability values, influencing powers of other persons' actions applied to the document are calculated.

A user' s influencing power is classified according to subjects, and the user's influencing power over a specific subject is primarily calculated as a sum of actions to documents included in the subject. Here, the "sum of actions" is obtained in such a manner as to calculate a probability of coinciding with a subject through automatic classification of all actions applied by users, and to correct the probability values by an action weight, thereby determining the sum of actions as a user's influencing power.

An algorithm for an action rank is implemented in such a manner as to sum up the results of digitization with respect to the types of actions described above. A weight calculation method according to each action is as follows.

( 1 ) Weight for Created Document

An influencing power of a document composed by a user is calculated by using weights of the user' s documents included in a specific subject and a total of other persons' influencing powers.

When a weight for a j^th document in relation to a specific subject "s" is expressed as W(s,d_j),

W(M_;.)=P(s,d_; ^.)x(αxP(d)+y#xR(d)+/xC(<0). The α, β, and Y represent weights for respective actions.

The "P(s,dj)" represents a probability that the j^th document coincides with the subject "s."

The "P(d)" represents a total of plus influencing powers up to date "d, " and is equal to "P' (d-1 )+p(d) , " wherein the "P' (d-1)" represents a total of plus influencing powers up to one day before date "d, " and the ^λλp(d)" represents a plus influencing power of another person on date "d." The ^λλR(d)" represents a total of duplicate influencing powers up to date "d, " and is equal to "R' (d-l)+r (d) ," wherein the "R' (d-1)" represents a total of duplicate influencing powers up to one day before date "d," and the ^λλr(d)" represents a duplicate influencing power of another person on date "d."

The "C(d)" represents a total of click influencing powers up to date "d, " and is equal to "C (d-1) +c (d) , " wherein the "C (d-1)" represents a total of click influencing powers up to one day before date "d, " and the "c (d) " represents a click influencing power of another person on date "d."

The weight of a document is basically influenced by the subject of the document, into which the document is classified through the automatic classification. Here, the "P(s,d_j)" is a probability value of a document according to each subject, which has been created through the automatic classification. The weight of a document is obtained by setting a probability value that the document is to coincide with a subject, to which the document belongs, as a basic value, and correcting the probability value by an influencing power of an action applied to the document by another user.

The influencing power of another user is roughly classified into three types: plus, duplicate, and click. This is calculated by adding a total of influencing powers of each action up to one day before date "d" and an influencing power of another person on date "d" once a day, in consideration of system resources and time. Finally, weights of actions described hereinafter in this document use weights in which a regression coefficient estimated through a simulation has been reflected.

(2) Weight by my plus action When a plus action is generated, a predetermined part of my influencing power is granted to a document author upon calculating a document weight. A predetermined part of the weight of the plus-action- subjected document in relation to a subject, to which the document belongs, is granted as my influencing power to me. A plus weight is determined by the following equation.

When a plus weight for an 1^th document belonging to a specific subject "s" is expressed as "W(s, Pi),"

The "P(s,di)" represents a probability that an 1^th document coincides with a subject "s."

The "AR(dχ,Ui)" represents an action ranking score of an owner of a document "d, " with respect to the subject "s," to which the document "d" belongs.

In the influencing power of each document according to each subject, a value obtained by multiplying a probability that the document belongs to the subject by the influencing power of the owner of the document is reflected.

It is assumed that a plurality of persons may perform plus actions with respect to one document, and in this case, plus influencing powers are identically distributed.

(3) Weight by My Duplicate Action

A duplicate action is calculated in the same manner as the plus. A duplicate weight is determined by the following equation. When a duplicate weight for an o^th document belonging to a specific subject "s" is expressed as "W(s, T₀),"

W(s,r_o)=P{s_td_o)xAR{d_otu_t) .

The "P(s, d_o)" represents a probability that an o^th document coincides with a subject "s."

The "AR (d_o, Ui)" represents an action ranking score of an owner of a document "d, " with respect to the subject "s," to which the document "d" belongs. (4) Click Weight by My Click Action

A click also is calculated in the same manner as the plus. However, it is expected that the plus has a far higher importance than the click. With respect to the click, it is necessary to prevent malicious click actions by employing a duplicate removal method (it is proposed that one user should be allowed to click only once a day with respect to each document).

When a click weight for an m^th document belonging to a specific subject "s" is expressed as "W(s, c_m) , "

W(s,c_m)=P(s,d_m)xAR(d_m,u_t) .

The "P(s,d_m)" represents a probability that an m^th document coincides with a subject "s."

The ^XλAR(d_m,Ui)" represents an action ranking score of an owner of a document ^Xλd, " with respect to the subject "s," to which the document "d" belongs.

A final action ranking is formulated into two types according to whether a document author exists or not, by combining three types of weights described above.

(1) Type of Document where the Author is Known (hereinafter, referred to as "type A")

Action rankings of type A are classified into a user ranking and a document ranking. A user's action ranking is determined by synthetically using a weight of a created document, a plus weight generated by my plus action, and a click weight generated by my click action. A user's action ranking based on type A is as follows.

The above equation implies that the ranking of a user "U_j" interested in a specific subject "s" gets higher, as the number of actions by other users with respect to all documents composed by the user increases

( ∑W(s,d_j) ) , and as the number of actions by the user

with respect to documents coinciding with the subject

increases ( ∑W(s,_Pι)+∑W(s,r₀)+∑W(s,c_m) ) .

1=1 o=l m=\

A document's action ranking is determined by merging the weight of a created document, a user's action ranking, and a system ranking. Based on type A, a document's action ranking is as follows. AR_A(s,d_j)=W(s,d_j)xAR_A(s,_Ui)

The above equation shows that a document's action ranking is obtained by multiplying the weight "W(s,dj)" of a created document by the document users' action scores "ARA(s, U_j) . " This enables the document's action ranking to get higher, as the number of actions by other users interested in a specific subject "s" with respect to document "d_j" is large, and as a user of the document is a guru in the subject.

(2) Type of Document where the Author is Unknown (hereinafter, referred to as "type B")

AR_B(s,d_t) =axP"+βxC"+χxSR+k

P": number of times of pluses

C": number of times of clicks

SR: system rank value k: constant

Since a document of type B has no information about a subject, only the number of times of actions is considered and used.

(3) Date Reflection A time period from the date when a document was created to the current date when an action ranking is calculated is digitalized, and an action ranking value is multiplied by the digitalized time period, so that more recent writing can have a higher ranking. When the "AR_A(s,d_j)" and the "AR_B(s,d_j)" are multiplied by "0.998^A (d_c-d_w) , " a graph as shown in FIG. 3 is obtained. However, a minimum time weight is "0.5", and from one year after a document is created, "0.5" is uniformly multiplied to reflect the time variant. Here, "d_c" represents the time when an action ranking is calculated, and "d_w" represents the time when a document was created.

Action ranks according to each type considering the final time weight is as follows.

A-type User Ranking

xO.999^Λ (d_c-d_w)

A-type Document Ranking

AR_A(s,d_J)=(W(S^₁)XAR_A(s,u_i))x0.999^(d_c-d_w)

B-type Document Ranking

AR_B(s,d_t)=(axP^*+βχC"+χxSR+k)χ0.999^Λ(d_c-d_w)

Upon a search, an application method is as follows.

A search also is classified into type A and type B, and the search is performed according to the system configuration shown in FIG. 4, as described below. While the B-type search is accomplished by deciding B-type action rankings according to the existing search method, the A-type search is accomplished by searching subject information that most closely coincides with a query, and then by combining the subject information with an existing ranking, differently from the existing search method. The following description will be given about a method of applying a search for documents and users in connection with type A.

( 1 ) Document Search

This is in a similar vein as the conventional search, but the internal search method thereof extracts a result obtained through combination of a keyword matching and a subject matching.

Search Algorithm

1) Find subjects by mapping multiple keywords to multiple subjects (a maximum of three)

-^ using a cosine coefficient between a query vector and a subject vector

2) Extract documents including the query vector as a set of resultants

-> using a keyword matching between queries and documents

3) Calculate a final ranking in reflection with an actio ranking

Here , A^R _A ^(dfo⁾ j subject value dj_: i document q: query ave: average value

AR,(*„<<,) α, β: action ranking score and SR' s weight

SR: system rank value (i.e., search engine's ranking score)

Therefore, which of actions considering even a relation between a query and a subject and a system rank is more weighted can be politically controlled by weights .

Therefore, the α is obtained by making reference to a class regression coefficient, and the β is obtained by means of a system regression coefficient.

A search query may be a compound query. According to query or subject classification, a result of the search may appear not as one subject, but as a set of subjects including multiple subjects. From among the result, only a maximum of three subjects having relatively higher cosine coefficients are extracted.

This enables the search reaction speed of an engine to be prevented from being reduced. Then, documents searched by the keyword matching are finally ranked through action ranks, and are retrieved.

(2) Person Search

This is similar to the document search, but is different from the document search in that the person search extracts only persons in a subject. Search Algorithm

1) Find a subject by mapping multiple keywords to multiple subjects

-^ using a cosine coefficient between a query vector and a subject vector 2) Extract users belonging to the mapped subject as a set of entire resultants

3) Calculate a final ranking in reflection with an action ranking according to a query

FIGs. 5 and 6 illustrate a list of documents and a list of users, that are provided by the aforementioned algorithm.

Industrial Applicability As can be seen from the foregoing, according to the present invention, documents and users that coincide with a specific query are exposed, and far superior search performance as compared with the existing simple keyword-based search ranking method is provided by means of user feedback information.

Claims

1. A method for granting a document ranking, the method comprising: a document composition step of composing a document by a user; a query input step of inputting a query by a searcher; a document weight extraction step of extracting a document weight, in consideration of both a probability that the document composed in the document composition step coincides with the query, and a sum of a plus influencing power, a duplicate influencing power, and a click influencing power up to a search date for the document; a user ranking extraction step of extracting a user ranking, from among a total sum of document weights which are caused by other persons' actions with respect to documents connected with the query among documents composed by the user, a total sum of plus weights by the user with respect to other persons connected with the query, a total sum of duplicate weights, and a total sum of click weights; and a document ranking extraction step of extracting a document ranking from the document weight and the user ranking.

2. The method as claimed in claim 1, wherein the document ranking is calculated by an equation,

wherein the "W(s,d_j)" represents a document weight, and the "AR_A(s,ui)" represents a user ranking.

3. The method as claimed in claim 2, wherein the document weight ^xvW(s,d_j)" is calculated by an equation, W(s,d_j)=P(s,d .)x(aχP(d)+β^χR(d)+γxC(d)) ,

wherein the "α", "β", and ^λλγ" represent weights for respective actions; the "P(s,d_j)" represents a probability that a j^th document coincides with a subject "s"; the "P(d)" represents a total of plus influencing powers up to date "d" and is equal to "P' (d-l)+p(d)," in which the "P'(d-l)" represents a total of plus influencing powers up to one day before date "d, " and the "p(d)" represents a plus influencing power of another person on date "d"; the "R(d)" represents a total of duplicate influencing powers up to date "d" and is equal to "R' (d- l)+r(d)," in which the "R'(d-l)" represents a total of duplicate influencing powers up to one day before date "d, " and the "r(d)" represents a duplicate influencing power of another person on date "d"; and the "C(d)" represents a total of click influencing powers up to date "d" and is equal to "C (d-l)+c(d)," in which the "C'(d-l)" represents a total of click influencing powers up to one day before date "d, " and the "c(d)" represents a click influencing power of another person on date "d."

4. The method as claimed in claim 2, wherein the user weight "AR_A(s,Ui)" is calculated by an equation,

AR_A{s,_Ul)=axYw{s,d^βfw(_S,_Pι)+γ^y{s,r₀)+δγW{s,c_m) , j=l I=I o=l m=\ wherein the document weight is defined by W(s,d_j)=P(s,d_j)x(axP(d)+β^χR(d)+rxC(d));

the plus weight is defined by

the duplicate weight is defined by

the click weight is defined by W(s,c_m)=P(s,d_m)xAR(d_m,_Ui) ,

wherein the "P(s,di)" represents a probability that an 1^th document coincides with a subject "s"; the "AR(di,ui)" represents an action ranking score of an owner of a document "d, " with respect to the subject "s," to which the document "d" belongs; the "P(s, d_o)" represents a probability that an o^th document coincides with a subject "s"; the "AR(d_o,Ui)" represents an action ranking score of an owner of a document "d, " with respect to the subject "s," to which the document "d" belongs; the "P(s,d_m)" represents a probability that an m^th document coincides with a subject "s"; and the "AR(d_m,ui)" represents an action ranking score of an owner of a document "d, " with respect to the subject "s," to which the document "d" belongs.

5. The method as claimed in any one of claims 1 to 4, wherein a time weight is applied to at least one of the user ranking and document ranking so that the corresponding ranking becomes lower as time passes.

6. The method as claimed in claim 5, wherein the time weight is defined by a formula, and the application means that the user ranking and document ranking are multiplied by the time weight, 0.998^Λ (d_c-d_w),

wherein the "d_c" represents a time when an action ranking is calculated, and the "d_w" represents a time when a document is created.

7. The method as claimed in any one of claims 1 to 6, wherein a final document ranking is calculated by applying a system ranking to the document ranking.

8. The method as claimed in claim 7, wherein the final document ranking "

" is calculated by an

equation,

wherein the AR^d^q) represents a final document

ranking for a query "q"; the "S_j" represents a j^th subject value; the "di" represents an i^th document; the "q" represents a query; the "ave" represents an average value;

the " represents a cosine value between a

subject vector ^λλS_j" and a query vector "q"; the " AR_n(S^d₁) " represents an A-type action rank

score of an i^th document of documents belonging to a j^th subject; the ^ΛΛα" and "β" represent an action ranking score and a SR' s weight; and the "SR" represents a system rank value (i.e., a search engine's ranking score).

9. The method as claimed in claim 8, comprising a document ranking extraction step of extracting a document ranking for a document of which an author is unknown, in which the document ranking for the document of which the author is unknown is extracted in consideration of influencing powers of a number of times of pulses and a number of times of clicks with respect to the document, and a system rank value.

10. The method as claimed in claim 9, wherein the document ranking "AR_B(s,d_j)" for the document of which the author is unknown is calculated by an equation,

wherein the P" represents a number of times of pluses, the C" represents a number of times of clicks, the SR represents a system rank value, and the k represents a constant.

11. The method as claimed in claim 10, wherein a time weight is applied to the document ranking of the document of which the author is unknown, so that the ranking becomes lower as time passes.

12. The method as claimed in claim 11, wherein the time weight is defined by a formula, and the application means that the document ranking of the document, of which the author is unknown, is multiplied by the time weight,

0.998^Λ(d_c-dj,

wherein the ^Xλd_c" represents a time when an action ranking is calculated, and the "d_w" represents a time when a document is created.

13. A method for granting a document ranking, the method comprising the steps of: calculating a weight for a first document according to a user' s action applied to the first document; calculating a first degree of matching between the calculated weight for the first document and a first keyword; and arranging the first document and a second document according to the first degree of matching and a second degree of matching when a query by the first keyword is generated, the second document having the second degree of matching equal to or greater than a predetermined value with respect to the first keyword.

14. The method as claimed in claim 13, wherein the weight for the first document is proportional to a ranking of a user who composes the first document.

15. The method as claimed in claim 13, wherein the weight for the first document is proportional to a number of times of bookmarks of the first document or a number of times of replies added to the first document, by a user reading the first document.

16. The method as claimed in claim 13, wherein the weight for the first document is proportional to a number of users who have read the first document.

17. The method as claimed in claim 13, further comprising a step of setting the degree of matching of the first document with respect to the first keyword to a higher value than the degree of matching of the second document, when the first and second documents include the same contents, and an original author of the same contents is an author of the first document.

18. The method as claimed in claim 13, wherein the weight for the first document is proportional to a temporal sequence of a user' s action generated with respect to the first document.

19. A computer-readable recording medium in which a program for implementing a method according to any one of claims 1 to 18 is recorded.