KR20110023308A - Method of searching web documents and system for performing the method - Google Patents

Abstract

PURPOSE: A web document search method and a system performing the same are provided to supply a better web document search result by reflecting the preference of the user. CONSTITUTION: A transaction analyzer(100) collects and analyzes web pages corresponding to a search key supplied from a user. A resource manager(120) manages information about a web document search process. A transaction processor analyzes the request of the user by combining a link based ranking scheme. The transaction processor classifies the action of the users according to the analysis result.

Description

METHOOD OF SEARCHING WEB DOCUMENTS AND SYSTEM FOR PERFORMING THE METHOD}

The present invention relates to a web document retrieval method and a system for performing the same. More particularly, the present invention relates to a collective intelligence-based web document retrieval method and a system for performing the same.

In general, an information retrieval system refers to a system that provides information requesters by quickly finding information suitable for a request from a database accumulated in an easy-to-find form by collecting, processing, and processing information or data that is expected to be needed by an information consumer.

In the past, a web portal site was constructed by using such an information retrieval system. The web portal site utilizes an information retrieval system to collect, process, and process information on a web site through the Internet, and provide the retrieved information to a user terminal, for example, a computer or a wireless terminal.

On the other hand, when the conventional information retrieval method is applied in a new field and terminology is rapidly increasing due to the development of information communication and technology, it is necessary to update the relevant information retrieval system from time to time as the new field or term is increased. In particular, the modification of the document DB in which the information on the huge documents is stored takes a lot of time and money, so it is difficult to carry out from time to time.

On the other hand, with the exponential growth of information available on the Internet, Internet users have to spend a lot of time every day looking for the necessary web pages out of a huge amount of Internet documents. Numerous web document search techniques are being developed to handle huge amounts of web pages, and web document search engines are becoming more complex. However, in the qualitative aspect of search, about half of the searched web pages are reported as unnecessary.

Although such a search system has been developed to find appropriate web pages that reflect user preferences, there are some issues that have not been solved yet. The biggest problem is that even though the search system can get numerous web pages on the Internet that reflect the user's preferences, it is still unsatisfactory to collect and analyze them due to the huge amount of web pages.

Therefore, the technical problem of the present invention has been made in view of this point, the object of the present invention is to reflect the user behavior and preferences in the search to adapt to the user of a large number of web pages on the Internet in order to obtain a better search results collective intelligence To provide a web document search method based on the above.

Another object of the present invention is to provide a web document search system for performing the web document search method described above.

In the web document search method according to an embodiment to realize the above object of the present invention, as a search keyword including a tag of web pages is input by a user, the storage server of the resource management layer stores the search keyword. And transmitting information about a user tag to a transaction processing layer, and the Foxsonomi manager of the transaction processing layer edits the tag and infers a relationship between the tags to analyze whether a user clicks on a web page. The indexing module of the transaction processing layer classifies the web page based on the tag, and the ranking module of the transaction analysis layer personalizes the page rank score to reflect the recorded behavior of users. Web belonging to same category using link based ranking algorithm Sorting the pages, and the web search interface of the transaction analysis layer includes displaying the ordered web document search results to the user.

In an embodiment of the present invention, the FoxMono manager assigns a unique ID to every web page clicked by a user, generates a tag tree, calculates the number of users using the same tag, and calculates the number of users calculated by an iterative program. Insert a high frequency tag into the tag database as the last category of a web page.

In an embodiment of the present invention, the page rank score is a web document search method, characterized in that defined by Equation 2 below:

[Equation 2]

(Where α is a damping factor, which is 0.85. V _i is a user-customized value, defined by Equation 3 below. The user's custom-value is defined by the user clicking on all pages linked to page i. It is determined by counting the number of clicks a user clicks on a web page that is linked to all pages and reflects the user's preference for the web page.)

&Quot; (3) "

(Where C _ji (u) is the number of times user u clicked from page j to page i, defined by Equation 4 below, where U is the total set of users:

&Quot; (4) "

(In Equation 2, S _ji is the number of mutual outlinks from page j when page j is linked to page i, and is defined by Equation 5 below. It is a number.)

 [Equation 5]

In order to realize the above object of the present invention, a web document retrieval system according to an embodiment collects and analyzes web pages corresponding to a search keyword provided from a user, and analyzes and displays a search result to the user. And a resource manager that stores and manages information about the web document retrieval process, and a combination of folksonomy and link-based ranking techniques to analyze user requests and classify user behaviors and interests according to the analyzed results. It includes a transaction processing unit provided to.

In an embodiment of the present invention, the transaction analyzer may include a web search interface for receiving a search keyword by a user and providing a web page corresponding to the search keyword to the user, a web crawler for collecting web pages; The ranking module may arrange the web pages belonging to the same category to be provided to the web search interface so that the web pages corresponding to the search keywords are provided to the user.

In an embodiment of the present invention, the resource manager may include a storage server including a folkson repository and a link information repository.

In an embodiment of the present invention, the transaction processing unit, a Foxnome manager for editing a tag corresponding to a web page, inferring relationships between the tags, and analyzing a user's behavior, and a web page based on the tag It may include an indexer module for classifying by category.

In an embodiment of the present invention, the folkson manager is a folkson editor for writing a tag input by the user corresponding to a specific web page to the web page, and the user to check whether or not the user clicks on the tagged web page The search analyzer may include a tag association extractor configured to generate a tag tree between a tag and a page by assigning an ID to the page as a specific web page is clicked by a user.

In an embodiment of the present invention, the indexer module may include: a syntactic analyzer for analyzing a language written on a foxsonized web page, a feature selector for selecting features of the language analyzed by the language analyzer, and A tag recognizer for recognizing a tag of a web page corresponding to the language selected by the feature selector, and a semantic classifier for classifying the meaning of the tag recognized by the tag recognizer and storing the semantic classifier in the resource manager. have.

In an embodiment of the present invention, the page rank score calculated by the ranking module may be defined by Equation 2 below.

[Equation 2]

(Where α is a damping factor, which is 0.85. V _i is a user-customized value, defined by Equation 3 below. The user's custom-value is defined by the user clicking on all pages linked to page i. It is determined by counting the number of clicks a user clicks on a web page that is linked to all pages and reflects the user's preference for the web page.)

&Quot; (3) "

(Where C _ji (u) is the number of times user u clicked from page j to page i, defined by Equation 4 below, where U is the total set of users:

&Quot; (4) "

(In Equation 2, S _ji is the number of mutual outlinks from page j when page j is linked to page i, and is defined by Equation 5 below. It is a number.)

 [Equation 5]

According to the web document retrieval method and a system for performing the same, a combination of folksonomy and link-based ranking technique based on collective intelligence is used to better reflect user behaviors and preferences in search to adapt to users among a large number of web pages on the Internet. Provide web document search results.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS The present invention will now be described in more detail with reference to the accompanying drawings. As the inventive concept allows for various changes and numerous embodiments, particular embodiments will be illustrated in the drawings and described in detail in the text. However, this is not intended to limit the present invention to the specific disclosed form, it should be understood to include all modifications, equivalents, and substitutes included in the spirit and scope of the present invention.

Like reference numerals are used for like elements in describing each drawing. In the accompanying drawings, the dimensions of the structures are shown in an enlarged scale than actual for clarity of the invention.

The terms first, second, etc. may be used to describe various components, but the components should not be limited by the terms. The terms are used only for the purpose of distinguishing one component from another. For example, without departing from the scope of the present invention, the first component may be referred to as the second component, and similarly, the second component may also be referred to as the first component. Singular expressions include plural expressions unless the context clearly indicates otherwise.

In this application, the terms "comprise" or "have" are intended to indicate that there is a feature, number, step, action, component, part, or combination thereof described on the specification, and one or more other features. It is to be understood that the present invention does not exclude the possibility of the presence or the addition of numbers, steps, operations, components, parts, or combinations thereof.

Also, unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. Terms such as those defined in the commonly used dictionaries should be construed as having meanings consistent with the meanings in the context of the related art and shall not be construed in ideal or excessively formal meanings unless expressly defined in this application. Do not.

The web document retrieval system according to the present invention is based on collective intelligence. Here, collective intelligence is getting more and more attention due to the arrival of new communication technologies. Collective intelligence means new insights created by a group of people combining actions, circumstances, and ideas.

Web document retrieval system based on collective intelligence ranks web pages according to user's preference through the combination of Folksonomy and Link. Here, folksonomy is a new classification technique for creating and managing tags for cooperatively describing and classifying content. This means a tag or label attached to each web page to describe and classify the content. Users put keywords called tags freely and subjectively to match the meaning of each page. Anyone can choose any word with a tag, and tag multiple pages on the same page.

Web 2.0 means a platform where nobody owns data and everyone can use it in any programming or Internet environment. In the meantime, Web sites have been unilaterally providing information and services like TV or radio, which is often referred to as media. Until now, it was not possible to move or utilize data posted to a website or serviced data. But once a Web 2.0 environment is built, you can move data around freely. If the representative service pointing to Web 1.0 is a web portal service, Web 2.0 means a platform. The service of the web portal site is not at your disposal, but the platform, Web 2.0, allows you to utilize the data as you want. In these Web 2.0, an increasing number of examples of using a tag to create a web document.

1 is a block diagram illustrating a web document search system according to the present invention.

Referring to FIG. 1, the structure of a web document retrieval system according to the present invention is based on a bus line framework. On both sides of the bus line framework 11, there is a management layer and a resource layer. In the present embodiment, the term hierarchical is a logically divided term, not a hardware term. Meanwhile, the bus line framework may be defined as a web page storage unit, a management layer may be defined as a management unit, and a resource layer may be defined as a resource storage unit. However, hereinafter, the term hierarchy will be used as it is.

A large amount of web page data is stored in the data bus layer.

The management hierarchy is provided with a folkson manager 12, a search engine 13 and a QA engine 14. The folksonomie 12 is responsible for analyzing and classifying a large amount of data. That is, the folkson manager 12 extracts the tag from the web document with the tag among the documents stored in the data bus 11 and assigns the extracted tag to the subject, and assigns the field using the folksonomi. The search engine 13 and the QA engine 14 process a user's request and return a search result to the user.

The resource layer includes a folkson storage 15 and a link information storage 16, and serves to store the corresponding information of the components of the management layer.

2 is a conceptual diagram illustrating an example in which users tag a web page.

Referring to FIG. 2, users were randomly divided into 'group A' and 'group B' each consisting of 10 members.

The three people in group A used "portal" as the tag for page B, and the other seven used the tag "news" for the same page. In this embodiment, the frequency of one tag per page is defined by the number of users who use the same tag on the same page, and the classification of the page is determined using the frequency of each tag assigned to the page. In other words, the tag with the highest frequency becomes the category of the page.

According to FIG. 2, the search system selects "news" as the category of page B because the frequency of the "news" tag is higher than the frequency of the "portal" tag. Similarly, the "news" and "music" tags are categories for page A and page C, respectively.

Table 1 shows the results of the page classification for FIG.

In this embodiment, it can be seen that the categories of page A and page B are 'news' which is the same category.

The most important advantage of Foxsonomy is that users can quickly search and easily categorize the associated web pages. Foxonomies are known to provide flat, non-hierarchical, shared terms. Although it has many useful features, it is a decisive problem when used in search. That is, the order of the retrieved pages depends on the popularity of the tag assigned to each page. As a result, Foxsonmi reflects the user's behavior well, while not fully reflecting their preferences. Therefore, high quality search engines need to consider additional techniques for accommodating user preferences all at once. Accordingly, the present invention utilizes one of the possible techniques, the link based ranking technique.

On the other hand, users have been able to obtain pages that meet user requests so far, but the order of pages obtained by users is simply the order of crawling. If the set of pages is large, users should refer to each search term from a lot of unnecessary content to actually select the pages they are interested in. To solve this problem, it is necessary to obtain the ranked results from the web search system. In general, the ranking techniques used in existing page recommendation systems are obtained through one of the categories of content based ranking and link based ranking.

The content-based ranking determination method analyzes information related to a user's preference and returns a page in order of score for a given search word. The statistics of this score are based on the content of all pages. Although most search engines continue to work this way, a little by considering the information that other people provide about the page, specifically who linked the page and what they said about the page, Can be improved. This improved method is called link-based ranking.

3 is a block diagram illustrating a web document search system according to an embodiment of the present invention.

Referring to FIG. 3, the web document retrieval system 100 according to an embodiment of the present invention includes a transaction analysis layer, a resources management layer, and a transaction process layer. In the present embodiment, the term hierarchical is a logically divided term, not a hardware term. Meanwhile, the transaction analysis layer may be defined as a transaction processor, the resource management layer may be defined as a resource manager, and the transaction processing layer may be defined as a transaction processor. However, hereinafter, the term hierarchy will be used as it is.

The transaction analysis layer is provided with a web search interface 112, a web crawler 114, and a ranking module 116 to collect and analyze web pages corresponding to a search keyword provided from a user, and to analyze the search results to the user. Is displayed.

The web search interface 112 receives a search keyword by the user and provides a web page corresponding to the search keyword to the user.

The web crawler 114 collects web pages existing on the web.

The ranking module 116 arranges web pages belonging to the same category and provides them to the web search interface so that the web pages corresponding to the search keyword are provided to the user.

The transaction analysis layer is provided with a search engine component and a QA engine component of the system structure described in FIG. That is, in the transaction analysis layer, information is collected and analyzed from the user, and in the resource management layer, the analysis result of the information is transmitted. In addition, an operation of displaying a search result to a user is performed in the transaction analysis layer.

The resource management layer is provided with a storage server 120 to manage and store information about a search process.

The storage server 120 is provided with a folkson storage and a link information storage described in FIG.

The transaction processing layer is provided with a folkson manager 130 and an indexer module 140, by combining folksonomy and link-based ranking techniques to analyze the user's request, and classify the user's behavior and interest according to the analysis result Provided to the resource management department.

The folksonography manager 130 includes a folkson editor 132, a tag association extractor 134, a user search analyzer 136, edit a tag corresponding to a web page, infer the relationship between the tags, Analyze user behavior.

The folkson editor 132 writes a tag input by a user in correspondence with a specific web page to the corresponding web page.

The tag association extractor 134 generates a tag tree between a tag and a page by assigning an ID to the corresponding page as a specific web page is clicked by a user.

The user search analyzer 136 checks whether the user clicks on the tagged web page.

The indexer module 140 includes a language analyzer 142, a feature selector 144, a tag recognizer 146, and a semantic classifier 148 to classify web pages based on the tag.

The language analyzer 142 analyzes the language written on the foxsonized web page.

The feature selector 144 selects features of the language analyzed by the language analyzer.

The tag recognizer 146 recognizes a tag of a web page corresponding to the language selected by the feature selector.

The semantic classifier 148 classifies the meanings of the tags recognized by the tag recognizer and stores them in the resource manager.

The transaction processing layer is equipped with a foxnomy management component. That is, the request of the users is analyzed in the transaction processing layer, and information about the actions or interests of the users is classified based on the analyzed result.

According to the present invention, by introducing a personalized link-based ranking technique that considers other web pages linked to the target web page, the limit of measuring the importance of the web page is increased by increasing the accuracy of the search.

In addition, according to the present invention, by introducing a personalized link-based ranking technique without depending on a specific data source such as a bookmark, it is possible to improve the accuracy of the web document retrieval system.

In addition, according to the present invention, since the web document retrieval system is designed based on the collective intelligence combining the folksonomy and the personalized link-based ranking technique, the user can quickly find related web pages in consideration of the user's behavior and preferences.

In addition, according to the present invention, it is possible to find associated web pages among the entire web pages of the Internet that accommodate the user's preferences.

4 is a conceptual diagram illustrating an overall process of web page ranking using a combination of a FoxNomi and a link based ranking technique according to the present invention.

3 and 4, as a user inputs a search keyword such as a tag of web pages, the search keyword is stored in the storage server 120 of the resource management layer. Information about the user tag is sent to the transaction processing layer.

The foxson manager 130 of the transaction processing layer is used to edit tags, infer relationships between tags, and analyze user behavior. In addition, the indexer module 140 of the transaction processing layer classifies web pages based on tags. Web pages classified with the same tags belong to the same category.

Subsequently, web pages belonging to the same category are sorted by the ranking module 116. The ranking module 116 sorts web pages belonging to the same category by using a link-based ranking algorithm.

Finally, the results of the ordered web pages are displayed to the users by the web search interface 112.

5 is a flowchart illustrating a web document retrieval method using a web page ranking technique using a combination of the FoxNomi and the link-based ranking technique according to the present invention.

3 and 5, the web crawler 114 collects web pages and stores the collected web pages (step S110).

The indexer module 140 then parses the syntax (step S120) and classifies the tag-based meaning (step S130). That is, the indexer module 140 classifies web pages based on tags. For example, web pages classified with the same tags belong to the same category.

Then, it is checked whether or not the collection of web pages by the web crawler 114 has ended (step S140). If it is checked that it is finished, it is terminated. If it is not finished, it feeds back to step S110.

On the other hand, it is checked whether or not a search keyword such as a tag of web pages by a user is input (step S210). As the search keyword is input in step S210, the search keyword is inputted to the storage server 120 of the resource management layer. ) Information about the user tag is transmitted to a transaction process layer.

Subsequently, the Foxsonomi manager 132 of the transaction processing layer edits the user tag and infers the relationship between the tags to analyze a tag relationship for analyzing the user's behavior (step S220).

Next, the ranking module 116 sorts the web pages belonging to the same category by using a personalized link-based ranking algorithm (step S230).

Then, the results of the web pages ordered by the personalized link-based ranking algorithm are displayed to the users (step S240).

In FIG. 5, a series of operations consisting of steps S110 to S140 and a series of operations consisting of steps S210 to S240 are performed in parallel with each other.

As described above, this embodiment proposes a collective intelligence web document retrieval system combining a folksonomi and a personalized link based page ranking structure. In other words, the web document retrieval system according to the present invention is designed based on collective intelligence combining a folksonomi and a personalized link based ranking technique. The above method allows the user to quickly find relevant web pages in consideration of the user's individual behavior and preferences.

Then, in the following, a description of the Foxsonomi algorithm and link-based ranking algorithm.

<Foxonomy Algorithm>

In the web document retrieval system according to the present invention, the Foxsonomi algorithm is used to classify the retrieved pages by analyzing the tags and the user's behavior in the retrieval system. An example of a pseudo-code for explaining a folksonomy process according to the present invention is as follows.

========================================

input: tag Ti, page Pj

make edge e (Ti, Pj) between Ti and Pj if Pj has Ti.

for each page Pj

fmax = f (Ti, Pj) // frequency of tag Ti assigned to Pj

for each tag Tk of Pj

delete edge e (Tk, Pj)

if (f (Tk, Pj)> fmax) T = Tk

end

make edge e (T, Pj)

end

========================================

According to the above described Foxson process, when a user enters a tag in the user interface of the web document search system, the web document search system records the tag and captures the user's action, for example, clicking on the web page (or web document). do.

Then, you assign a unique ID to every web page that the user clicks, then build a tag tree. In the tag tree, the tags Ti and page Pj become parent and child nodes, respectively.

The web document retrieval system then calculates the number of users using the same tag.

Finally, the highest frequency tag calculated by the recursive program is inserted into the tag database as the last category of the web page. If the web pages are of the same category, then these web pages belong to the same category.

In such a folksonomy algorithm, if the total number of web pages is determined to be m and the total number of tags assigned to the web page is determined to be n, then the time complexity of the folksonomy algorithm is O (mn). Here, the time complexity is a time required according to the amount of data to be processed (indicated by N or n), which represents a proportional time rather than an absolute time, and is expressed as O (f (n)).

<Personalized Link-based Ranking Algorithm>

The personalized link based ranking algorithm according to the present invention is included in the ranking module 116 shown in FIG. In order to produce high quality search results, link-based ranking algorithms generally assign a score to each page indicating how relevant each page is. The relevance of each page is calculated from the number of other related pages linked to that page and the number of links to other pages.

Assume the relevance of a page as a PageRank score. The PageRank algorithm was developed by the founder of Google, and a change in this concept is currently used in very large search engines. In theory, pagerank is calculated as the probability that someone will randomly click on a link to reach a page. If the page contains links from other popular pages (inbound), it is more likely that someone accidentally visits the page. To capture this, the page rank uses a damping factor α that gives the user an opportunity to continually click on a link on each page. The value of α is determined by analyzing the behavior of the user clicking on the web page from another linked page. In general, α is assigned to 0.85. In general, the page rank score PRi of the web page i is calculated by Equation 1 below.

Here, α is a damping factor, which is 0.85. S _ji is the number of mutual outlinks from page j when page j is linked to page i. If there is no link from j to i, S _ji is zero. n is the total number of pages.

Meanwhile, in the present invention, the link-based ranking algorithm is used to reflect the recorded behavior of users.

Equation 1, which calculates the page rank score, is modified to be used in the link-based ranking algorithm as shown in Equation 2 below to reflect the recorded user's individual behavior. The time complexity of the clearly modified page rank algorithm is O (n2 + nq).

In Equation 2, PR _i is a personalized PageRank score. α is a damping factor, which is 0.85. V _i is a user-customized value for the customized search, and is defined by Equation 3 below. The user's custom value reflects the user's preference for the web page. The user's custom value is determined by counting the number of clicks by the user on all pages linked to page i and the number of clicks by the user on web pages linked to all pages.

Here, C _ji (u) is the number of times user u clicks from page j to page i, and is defined by Equation 4 below, and U is a set of all users.

In Equation 2, S _ji is the number of mutual outlinks from page j when page j is connected to page i, and is defined by Equation 5 below. If there is no link from j to i, Sji is zero. n is the total number of pages.

In order to prove the practical validity of the search system according to the present invention, a series of experiments were conducted to search web pages. Hereinafter, a search experiment of a web page according to the present invention will be described.

6 shows a result of searching a page with a search word news in a system implemented as a prototype according to the present invention.

Referring to FIG. 6, in this embodiment, page rank scores are calculated and arranged in order using a link based ranking technique. You can see that there are five pages that match the listed search terms. The score of each web page is displayed at the end of the page title.

That is, the page rank score of the web page associated with <rihanna-Search results for rihanna-CNN.com> was 0.15218, and the page rank score of the web page associated with <Commentaries: News & Videos about Commentaries> was 0.12945. Displayed. In addition, the page rank score of the web page related to <moos-Search results for moos-CNN.com> was 0.11017, and the page rank score of the web page related to <china-Search results for china-CNN.com>. Is displayed as 0.10100, and the page rank score of the web page related to <Benazir Bhutto: News & Videos about Benazir Bhutt> is 0.09598.

Users can identify and delete irrelevant pages on the search results screen shown in FIG. 6.

You can see that the next column of information related to each web page displays the link information to the web page.

In the search experiment according to the present invention, web pages were added to the web document search system, and Table 2 shows the added results.

Table 2 shows the results of the web pages added by the Foxson algorithm to the system.

Referring to Table 2, for example, the web page P1 is matched with 56 tags <news>, 11 tags <city>, and 3 tags <trip>, and the web page P1 is matched. The categories of were categorized into news categories.

7 shows the calculation results of the link structure and page rank.

Referring to FIG. 7, additional three pages, that is, P6, P7, and P8, which are connected to the previous five pages P1, P2, P3, P4, and P5, have been added to the calculation.

Three pages P6, P7, and P8 added in this experiment are connected to page P5, and the page rank scores of pages P6, P7, and P8 are all set to a value of 0.1. In this example, using the scores of pages P6, P7, P8 and using the personalized PageRank algorithm, 0.09598 could be obtained as the personalized PageRank score of page P5. Personalized PageRank scores of the other pages P1, P2, P3, P4 are also shown in FIG.

That is, using the personalized page rank algorithm, 0.15218, 0.12945, 0.11017, and 0.10100 were obtained as the personalized page rank scores of pages P1, P2, P3, and P4, respectively.

Tables 3 and 4 show details of the page rank calculation in this experiment.

Table 3 shows how to calculate the number of mutual outlink (S _ji ).

Referring to Table 3, the first column shows web pages linked to the pages of the search result list of FIG. For example, pages P2, P3, P4, and P5 are all linked to page P1.

In this embodiment, in order to calculate the value of the number S _ji of the mutual outlinks, it is necessary to calculate the number of users clicking on each link page. Here, the number S _ji of mutual outlinks is the number of mutual outlinks from page j when page j is linked to page i. If there is no link from j to i, S _ji is zero.

In Table 3, the total number of times a user clicks on page P1 from four pages P2, P3, P4, and P5 is 288. The total number of user clicks on P2, P3, P4, and P5 are 126, 57, 60, and 45, respectively. Users didn't click on P6, P7, or P8.

In the present embodiment, these values can be used to calculate S21, S31, S41 and S51.

Table 4 shows the final PageRank score (PR _i ) calculated according to the present invention.

Referring to Table 4, for the page P1, since the damping factor α is 0.85 and the user fit value is 15/41, 0.15218 is calculated as the personalized page rank score PR _i . In addition, for the page P2, since the damping factor α is 0.85 and the user fit value is 11/41, 0.12945 was calculated as the personalized page rank score PR _i . In addition, for the page P3, since the damping factor α is 0.85 and the user fit value is 7/41, 0.11017 is calculated as the personalized page rank score PR _i . In addition, for the page P4, since the damping factor α is 0.85 and the user fit value is 5/41, 0.10100 is calculated as the personalized page rank score PR _i . In addition, for the page P5, since the damping factor α is 0.85 and the user fit value is 3/41, 0.09598 was calculated as the personalized page rank score PR _i .

Although described above with reference to the embodiments, those skilled in the art can be variously modified and changed within the scope of the invention without departing from the spirit and scope of the invention described in the claims below. I can understand.

As described above, according to the present invention, by implementing a search system based on collective intelligence, it is possible to achieve a high quality search. In other words, by combining the folksonomy and link-based rank structure, it is possible to implement a search system based on collective intelligence that reflects both user behavior and user preference in search. In particular, by using Foxsonomi, a new classification method that combines users' preferences and thoughts with Web 2.0, and link-based ranking technique that supports personalization along with user behavior, the search effect can be improved.

1 is a block diagram illustrating a web document search system according to the present invention.

2 is a conceptual diagram illustrating an example in which users tag a web page.

3 is a block diagram illustrating a web document search system according to an embodiment of the present invention.

4 is a conceptual diagram illustrating an overall process of web page ranking using a combination of a FoxNomi and a link based ranking technique according to the present invention.

5 is a flowchart illustrating a web document retrieval method using a web page ranking technique using a combination of the FoxNomi and the link-based ranking technique according to the present invention.

FIG. 6 shows a result of searching a page by searching for news in a prototype system.

7 shows the calculation results of the link structure and page rank.

<Description of the symbols for the main parts of the drawings>

112: Web Search Interface 114: Web Chlorella

116 ranking module 120 storage server

130: Foxson manager 132: Foxson editor

134: Tag Association Extractor 136: User Search Analyzer

140: indexer module 142: language analyzer

144: feature selector 146: tag recognizer

148: semantic classifier

Claims (11)

When a search keyword including a tag of web pages is input by a user, the storage server of a resource management layer stores the search keyword and transmits information about the user tag to a transaction processing layer; The foxson manager of the transaction processing layer editing the tag and inferring a relationship between the tags to analyze whether a user clicks on a web page; The indexer module of the transaction processing layer classifies a web page based on the tag; The ranking module of the transaction analysis layer includes: arranging web pages belonging to the same category using a personalized link based ranking algorithm that calculates a page rank score to reflect the recorded behavior of users; And And the web search interface of the transaction analysis layer comprises presenting the ordered web document search results to a user. The method of claim 1, wherein the folkson manager, Give every web page the user clicked a unique ID, then generate a tag tree, Count the number of people who used the same tag, A method of retrieving a web document, comprising inserting the highest frequency tag calculated by the recursive program into the tag database as the last category of the web page. The method of claim 2, wherein in the tag tree, the tag is defined as a parent node, and the web page corresponding to the tag is defined as a child node. The method of claim 1, wherein the page rank score is defined by Equation 2 below: [Equation 2]

(Where α is a damping factor, which is 0.85. V _i is a user-customized value, defined by Equation 3 below. The user's custom-value is defined by the user clicking on all pages linked to page i. It is determined by counting the number of clicks a user clicks on a web page that is linked to all pages and reflects the user's preference for the web page.) &Quot; (3) &quot;

(Where C _ji (u) is the number of times user u clicked from page j to page i, defined by Equation 4 below, where U is the total set of users: &Quot; (4) &quot;

(Equation 1, S _ji is the number of mutual outlinks from page j when page j is linked to page i, and is defined by Equation 5 below. It is a number.)  [Equation 5]

A transaction analyzer configured to collect and analyze web pages corresponding to the search keyword provided from the user, and display the search result to the user; A resource manager to store and manage information about a web document search process; And And a transaction processing unit that combines a folksonomy and a link-based ranking technique to analyze the requests of users and classifies the behaviors and interests of the users according to the analyzed results and provides them to the resource manager. The method of claim 5, wherein the transaction analysis unit, A web search interface for receiving a search keyword by a user and providing a web page corresponding to the search keyword to the user; A web crawler that collects web pages; And And a ranking module for arranging web pages belonging to the same category and providing them to the web search interface such that web pages corresponding to the search keyword are provided to a user. The method of claim 5, wherein the resource management unit, A web document retrieval system comprising a storage server comprising a folkson repository and a link information repository. The method of claim 5, wherein the transaction processing unit, A FoxNomi manager for editing a tag corresponding to a web page, inferring a relationship between the tags, and analyzing a user's behavior; And And an indexer module for classifying web pages into categories based on the tag. The method of claim 8, wherein the folksonist A foxson editor for writing a tag input by a user in correspondence to a specific web page to the web page; A user search analyzer for checking whether a user clicks on a tagged web page by a user; And And a tag association extractor for generating a tag tree between a tag and a page by assigning an ID to the page as a specific web page is clicked by a user. The method of claim 9, wherein the indexer module A syntactic analyzer for analyzing the language written on the foxsonized web page; A feature selector for selecting a feature of the language analyzed by the language analyzer; A tag recognizer for recognizing a tag of a web page corresponding to a language selected by the feature selector; And And a semantic classifier for classifying the meanings of the tags recognized by the tag recognizer and storing the semantic classifiers in the resource management unit. The method of claim 6, wherein the page rank score calculated by the ranking module, Web document retrieval system, characterized by the following equation (2): [Equation 2]

(Where α is a damping factor, which is 0.85. V _i is a user-customized value, defined by Equation 3 below. The user's custom-value is defined by the user clicking on all pages linked to page i. It is determined by counting the number of clicks a user clicks on a web page that is linked to all pages and reflects the user's preference for the web page.) &Quot; (3) &quot;

(Where C _ji (u) is the number of times user u clicked from page j to page i, defined by Equation 4 below, where U is the total set of users: &Quot; (4) &quot;

(In Equation 2, S _ji is the number of mutual outlinks from page j when page j is linked to page i, and is defined by Equation 5 below. It is a number.)  [Equation 5]

Images