KR101543780B1 - System and method for expert search by dynamic profile and social network reliability - Google Patents

Abstract

The present invention relates to an expert search system and method using a dynamic user profile and a social network reliability, wherein a dynamic user profile management unit manages a dynamic user profile based on a number of documents created on a predetermined date range, a preference for a document, The reliability score of the social network for the user is calculated by the reputation score given to the network path between the users connected to the social network, The expert search unit compares the dynamic user profiles corresponding to the search query, adds the reliability scores to the compared values, and searches the experts by the order of similarity. According to the present invention, instead of the existing profile of the user, a dynamic user profile is periodically generated through analysis of recent activities, and an expert is searched in consideration of a user's reputation score, thereby ensuring accuracy and reliability of expert search, By using the generated dynamic user profile, users who are connected to the social network are searched for preferentially, thereby reducing the average search time cost.

Description

TECHNICAL FIELD [0001] The present invention relates to a system and a method for searching an expert using a dynamic user profile and a reliability of a social network,

The present invention relates to an expert search system and method, and more particularly, to a dynamic user profile analyzing a user's recent activity information and a dynamic user profile searching for an expert in consideration of reliability by evaluation in a social network and a reliability of a social network And more particularly, to a system and a method for searching experts using the same.

Due to the complexity and diversity of society, the information circulated is changing very rapidly and the needs of users are also changing. In addition, as the desire of the individual becomes stronger, the social network service (social network service) in which members of various societies express their own values online is spreading. The initial social network service has been used for socializing and entertainment purposes, but recently it has been increasingly used for information sharing and searching. Social networks offer a variety of services such as Facebook, Twitte, and Filckr, which promote communication among users while sharing information, and Linkedin, which focuses on networking and exchanges. Because these services generate and share various information centered on networking relationships among a large number of people, various types of information are provided indiscreetly. In this situation, untested information generated from multiple users is being shared. In addition, it is difficult to acquire high quality information due to restriction of contents or duplicated contents. Therefore, it is required to provide credibility of information shared and distributed through a social network and a technique for providing only meaningful information.

Thus, there is a need for techniques that can search for experts who can provide high quality information for a search request that a user desires. However, it is difficult for a user to search for a user (expert) who searches a large number of users existing in a social network and provides accurate information. In addition, we can not guarantee that users connected by social networks in social networks are experts who can provide high quality information. Accordingly, an expert among the users connected to the social network is usually identified through profile analysis of the user. Existing Expert Search is a way for all social network users to simply compare keywords of interest in their profiles. For example, when a user makes a search request, the requested query is compared to the profiles of all social network users. The higher the score of the requested keyword matches the interest keyword of the profile, the higher the score, and the ranking based on the score is provided as a result.

On the other hand, the expert search technique searches for an acquaintance who can best answer an inquiry in a social network. A user creates a profile when subscribing to various social network services. Each social network service consists of a set of friends, friends and friends' profiles, and tags included in the profile. Using this profile information, a user who matches the user's query is searched by an expert. F. Duchateau said, "Who can answer the question?" In this paper, we proposed a basic strategy, a cluster strategy, and a tree strategy to improve the accuracy of expert search performance. The basic strategy is to simply compare the set of interest tags contained in the query and profile. This technique takes very little time, but it does not search for synonyms and similar tags, so accuracy is reduced. The cluster strategy is a technique that integrates similar tags among the tags of the acquaintances and compares the queries with the clusters extended through ontologies. However, there is an advantage in that the accuracy of matching can be increased while expanding similar words and the like to the cluster, although there is a disadvantage that the time cost is increased because the clusters are formed by comparing the similarity degree of each tag and the cluster is expanded through the ontology. The tree strategy is a technique of calculating the similarity between the query tree and the tag tree of the acquaintance by constructing the query and the acquaintance tags as a tree. The degree of similarity is calculated according to how much the query tree and the tag tree share the common ancestor node. Tree strategy has a disadvantage in that there is a difference in search speed depending on the tree to be used, accuracy is lower than that of a cluster strategy, and time cost is less than a cluster strategy.

These existing expert search techniques are to search for experts by analyzing the interests of the profiles created by the user when they first sign up. This can lead to problems with accuracy and up-to-date if the user's recent interests have changed because the existing user's profile does not know whether the profile actually matches or has changed with the interests of the end user. As such, it is difficult to determine whether a user has actually created a profile of interest in a profile or whether a recent interest has changed, so it is difficult to search for an expert in a conventional manner. To solve these problems, techniques for analyzing user activity information are being studied. D. Horowitz proposed a technique for searching social network users who are most likely to answer a specific search query in a paper called "The Anatomy of a Large-Scale Social Search Engine." In addition to the analysis, the user to be selected as the expert compares the search query with the search query in consideration of the frequency of the answers to the answers previously left by the user, the length of the message, the speed of the answer, etc. Also, To improve the efficiency of the search, but it does not take into account the relationships among users of the social network, thus reducing reliability.

A. Bozzon, in his article entitled "Choosing the Right Crowd: Expert Finding in Social Networks," has proposed a technique for analyzing the user's recent activity information instead of the user profile, We focus on the relationship among users through social information analysis rather than processing performance for expert search. In other words, when a user's simple friend relationship is connected in a social network in a real world, it is determined that the user is a simple personal relationship and excluded from a user who can provide it with professional knowledge. This can reduce search costs. However, when analyzing recent activities, reliability is degraded because it does not take into account the reputation of documents and users.

In this way, various services are provided as social network services expressing individual opinions and values are developed. Therefore, when there is a problem of sharing unreliable information in this process, searching for an expert who provides reliable information An improved technique is needed.

Korean Patent Publication No. 10-2013-0045425 (Publication date 2013.05.06.) Korean Patent Laid-Open Publication No. 10-2013-0098511 (Publication date 2013.09.05.)

SUMMARY OF THE INVENTION Accordingly, the present invention has been made to solve the above-mentioned problems occurring in the prior art, and it is an object of the present invention to provide a method and system for updating a dynamic user profile based on recent activities in a social network, And to provide a system and method for searching experts using a dynamic user profile and social network reliability.

Specifically, the present invention periodically detects a user's activity and updates a dynamic user profile so that the profile is updated. This dynamic user profile can be compared with a search query to improve the accuracy of the results. The generated dynamic user profile can be compared with the extended search query through the ontology to improve the accuracy of the result. In other words, the search range is extended to words having similar meaning to produce more improved results.

Here, due to the nature of the social network service, information is generated and spread indiscriminately, so that inaccurate or malicious information can be shared among the users. In order to solve these problems, not only the contents of the articles left by the user but also the comments are analyzed and the reputation of the articles is evaluated to remove the users including the malicious contents in advance. Therefore, users can search for users who provide high-quality information by using reliability scores of each other.

In addition, search for friend relationship users for efficient expert search can reduce search time cost.

According to another aspect of the present invention, there is provided a system for searching for a dynamic user using a dynamic user profile and a social network reliability, the system comprising: A dynamic user profile manager for generating and updating a dynamic user profile by calculating expertise scores; A social network reliability calculation unit for calculating a reliability score of a social network for a user with a reputation score assigned to a network path between users connected to the social network; And an expert search unit for comparing the dynamic user profiles corresponding to the search query, adding the reliability scores to the compared values, and searching for experts according to the similarity rank.

Here, the dynamic user profile management unit may include a social information collection unit for collecting the document, social information including scrap, recommendation, and comment on the document, and the number of participants; A social information analyzer for extracting keywords by using a text mining technique in the document, calculating the preferences by analyzing scraps, recommendations, and comments on the documents, and assigning weights according to the number of participants; And a dynamic user profile generation unit for calculating the expertise score with the preference for the keyword and the weight, and generating a new dynamic user profile with keyword list, expertise score, and keyword frequency.

According to another aspect of the present invention, there is provided a method for searching a dynamic user profile using a dynamic user profile and a social network reliability, comprising the steps of: (a) Generating and updating a dynamic user profile by calculating expertise scores with given weights; (b) calculating, by the social network reliability calculation unit, a reliability score of the social network for the user with a reputation score assigned to a network path between users connected to the social network; And (c) the expert search unit compares the dynamic user profile corresponding to the search query, adding the reliability score to the compared value, and searching for experts according to the order of similarity.

The step (a) includes the steps of: collecting the document, scrap, recommendation and comment on the document, and social information including the number of participants; Extracting keywords from the document using a text mining technique, calculating the preferences by analyzing scraps, recommendations, and comments on the documents, and assigning weights according to the number of participants; And calculating the expertise score with the preference for the keyword and the weight, and generating a new dynamic user profile with the keyword-specific list, the expertise score, and the keyword frequency.

The step (b) comprises: collecting network paths between users; Selecting a specific user and users having a network path among the users; Collecting all network paths existing between the specific user and selected users; Calculating an average value of reputation scores in all the network paths; And calculating a reliability score of the specific user by dividing the weight assigned to each network path by the sum of average reputation scores of all paths. At this time, if there is no other user among the users, the reputation score is calculated as the reliability score.

The step (c) includes: generating a search query cluster expanded with synonyms, synonyms, and dialects using an ontology corresponding to the search query; Collecting a list of users directly connected to the search query user; Comparing the search query cluster with a dynamic user profile of the directly connected users; Calculating a degree of similarity by adding a reliability score to the compared value; Generating a similarity score list in descending order of the degree of similarity; And if the last rank of the degree of similarity is equal to or greater than a threshold value, the ranking of all experts above the threshold is determined and output as a result, and if the last rank of the degree of similarity is less than the threshold value, . At this time, if there is no synonym, synonym, or dialect, the search query is used as it is.

As described above, according to the expert search system and method using the dynamic user profile and the social network reliability according to the present invention, a dynamic user profile is periodically generated by analyzing the recent activity instead of the user's existing profile, By searching experts based on scores, it is possible to guarantee the accuracy and reliability of expert search.

In addition, according to the present invention, since the user connected to the social network is preferentially searched using the dynamic user profile generated in advance, the average search time cost can be reduced.

1 is a conceptual diagram of an expert search system using a dynamic user profile and a social network reliability according to an exemplary embodiment of the present invention.
2 is a conceptual diagram illustrating a process of generating a dynamic user profile according to an embodiment of the present invention.
3 is a diagram of a generated dynamic user profile according to one embodiment of the present invention.
4 is a flowchart illustrating a dynamic user profile creation process according to an exemplary embodiment of the present invention.
FIG. 5 is a diagram for creating a trust network among users according to an exemplary embodiment of the present invention.
6 is a flowchart illustrating a reliability evaluation process of a social relationship according to an embodiment of the present invention.
FIG. 7 is a diagram comparing the similarity between a query and a user profile according to an embodiment of the present invention.
FIG. 8 is a flowchart illustrating an expert search process according to an exemplary embodiment of the present invention.

The system and method for expert search using the dynamic user profile and social network credibility of the present invention can be used to analyze a user's recent activity without analyzing an existing static profile and to create a dynamic user profile and also to filter malicious users Search for experts using user reputation information. Since the dynamic user profile and the reputation information are used to search for a user who can provide a professional response, high quality information with high accuracy and reliability can be provided.

To this end, the present invention provides a dynamic user profile management system comprising: a dynamic user profile manager for generating and updating a dynamic user profile by calculating a professionalism score based on a number of documents created on a date range, a preference degree for a document, A social network reliability calculation unit for calculating a reliability score of a social network for a user with a reputation score assigned to a network path between users connected to the social network, a dynamic network reliability calculating unit for comparing the dynamic user profile corresponding to the search query, And an expert search unit for searching for experts according to the order of similarity by combining reliability scores.

Here, the dynamic user profile management unit includes a social information collection unit for collecting documents, scraps, recommendations, and comments on the documents, and social information including the number of participants, keywords using a text mining technique, A social information analysis unit for analyzing scraps, recommendations, and comments, calculating preferences, and assigning weights according to the number of participants; calculating expertise scores based on preferences and weights of keywords; And a dynamic user profile generating unit for generating a new dynamic user profile with a keyword frequency.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, a system and method for searching a professional using a dynamic user profile and a social network reliability according to the present invention will be described in detail with reference to the accompanying drawings.

1 is a conceptual diagram of an expert search system using a dynamic user profile and a social network reliability according to an exemplary embodiment of the present invention.

Referring to FIG. 1, in order to search for an expert, a recent dynamic activity profile of each user is collected and analyzed in a social network, and a new dynamic user profile reflecting a recent interest is generated. To increase the efficiency of real-time search, the server periodically detects the user's activity and generates a dynamic user profile. For the query requested by the user (queryer), an extended cluster is created with a query word and a similar word to guarantee the accuracy of the query result. Expert search compares query clusters with dynamic user profiles to determine expertise. We also consider trust between users in a social network to determine their expertise with trusted users. Then, ranking according to the degree of expertise, and providing a reliable and accurate user to the user (inquirer) as a result.

Social network users create their own profiles when they subscribe to the service. However, most users do not constantly update their profiles in social network services. However, user interest and expertise can be continuously changed. In this situation, users who do not have real expertise can be searched by a professional using only the user's profile. To solve this problem, the present invention analyzes a recent activity to generate a new dynamic user profile. The process of creating a dynamic user profile will now be described.

2 is a conceptual diagram illustrating a process of generating a dynamic user profile according to an embodiment of the present invention.

Referring to FIG. 2, a user of a social network performs various activities while using a service, and generates or shares specific information. Such information will remain on social networks in the form of articles, shared posts, interests, scraps, recommendations, and comments. Therefore, a new dynamic user profile is generated by analyzing the information left recently. First, in the social information gathering stage, after the documents related to the user are collected, the contents are divided into the contents, scraps, recommendations, comments, and the number of users participating in the document. Based on the collected information, the social information analysis stage generates preference through each document analysis. In the process of social information analysis, keywords are extracted using a text mining technique, and the preference of the document is calculated by the number of scraps, recommendation number, and comment analysis. The more people who participate in the document, the more popular and reliable the information, the more weight it gives. In the dynamic user profile creation step, the expertise score is given to each keyword in consideration of the preference and weight of the document. Since the same keywords may exist in different documents, the same keywords calculate the average and frequency of each expertise score. Upon completion of the recent activity analysis, a new dynamic user profile is created with keyword list, expertise score, and keyword frequency. Here, the functions performed in each step are performed in the social information collecting unit 1, the social information analyzing unit 2, and the dynamic user profile generating unit 3, respectively.

을 수집하고 키워드를 추출한다. 사용자가 작성한 문서의 스크랩, 추천 수, 댓글을 분석하여 해당 문서에 대한 선호도를 계산한다. 추출된 키워드에 대해 선호도를 고려하여 키워드별 전문성

를 계산한다. (식 1)은 문서

에서 전문성

을 계산하는 식이다. 이때,

는 문서

에 대한 가중치,

는 문서

에 대한 선호도,

는 사용자가 작성한 문서의 수,

는 상수이다. 각 문서를 분석하면 다수의 키워드가 추출되며 각 문서에 대한 선호도가 상이할 수 있다. 따라서 각 문서를 대표하는 키워드에 대해 선호도를 적용하고 사용자의 최근 활동량에 따라 전문성을 증감시킨다. 최근 활동량은 사용자가 최근에 남긴 글의 수를 의미하므로 글의 수가 적으면 적을수록 사용자를 평가하는 객관성이 떨어지게 되므로 각 문서마다 점수를 차감한다.
Docs shared by social network services over a recent period to analyze your recent activity

And extracts the keywords. Analyze the scrap, recommendation count, and comments of the user-created document and calculate the preference for the document. Keyword-specific expertise based on preference for extracted keywords

. (Equation 1)

Expertise in

. At this time,

Document

Weight,

Document

Preference for,

The number of documents created by the user,

Is a constant. Analyzing each document can extract a large number of keywords and may have different preferences for each document. Therefore, preferences are applied to the keywords representing each document and the expertise is increased or decreased according to the recent activity of the user. Since the recent activity amount means the number of articles that the user has left recently, the smaller the number of articles, the lower the objectivity to evaluate the user. Therefore, the score is deducted for each document.

--- (식 1)

--- (1)

에 대한 가중치

을 계산하는 식이다. 이때,

는 문서

에 참여한 사용자의 수,

는 상수이다. (식 2)에서

값보다 많은 사용자가 관심을 갖는 문서는 신뢰할 수 있으며 객관성이 있다고 판단한다. 즉, 특정 문서를 많은 사용자들이 읽거나 댓글을 생성할 경우 높은 가중치를 부여한다.
Many users share or refer to a document as trustworthy. Therefore, in the present invention, weights are given in consideration of the number of users participating in the document. (Equation 2)

Weight for

. At this time,

Document

The number of users who participated in,

Is a constant. (Equation 2)

Documents that are of interest to more than the user's value are considered reliable and objective. That is, when a particular document is read by many users or generates a comment, it gives a high weight.

--- (식 2)

--- (Equation 2)

에 대한 선호도는 스크랩 수, 추천 수, 댓글을 분석하여 (식 3)와 같이 계산한다. 이때,

는 상수,

는

에 대한 스크랩 수,

는

에 대한 추천 수이다. 또한,

는

에 대한 평판도로 긍정, 부정, 중립 여부에 따라 1, 0, 0.5의 점수를 부여한다. 만약

가 상수

보다 큰 값을 갖는 경우에는 1을 부여한다. 예를 들어, Facebook의 경우 사용자가 남긴 글에 다른 사용자들이 댓글을 작성할 수 있다. 이러한 상황에서 사용자가 작성한 특정 글에 다른 사용자들은 긍정, 부정, 중립과 같은 댓글 남길 수 있다. 이러한 댓글들은 글의 선호도로 표현될 수 있으며 악의적인 정보를 가진 글이나 사용자를 검색 결과 대상에서 제외할 수 있게 된다.
When a user gives a positive comment or recommendation to a specific document, the document can be judged to be professional. Therefore,

The number of scraps, the number of recommendations, and comments are analyzed and calculated as shown in (Equation 3). At this time,

Is a constant,

The

For scrap numbers,

The

. Also,

The

A score of 1, 0, or 0.5 is awarded depending on whether you are positive, negative, or neutral. if

Constant

If it has a larger value, 1 is given. For example, on Facebook, other users can comment on posts they leave. In this situation, other users can leave comments such as positive, negative, and neutral in a specific article written by the user. These comments can be expressed in the preference of the article and exclude articles or users with malicious information from the search results.

--- (식 3)

--- (Equation 3)

에서 추출된 키워드

에 대한 전문성

은

에 부여된

을 동일하게 부여한다. 사용자의 전문성을 판단하기 위해 키워드

에 대한 전문성

은 사용자가 작성한 문서에서 추출한 동일 키워드에 대한 전문성

의 평균 값으로 부여한다. 사용자가 작성한 문서를 통해 추출한 키워드에 대한 전문성

을 이용하여 사용자 프로필을 생성한다. 이러한 프로필은 사용자가 작성한 최근 문서를 이용하여 주기적으로 갱신한다.
document

Keywords extracted from

Expertise in

silver

Given to

. To determine your expertise,

Expertise in

Is the expertise of the same keyword extracted from the user-created document

As shown in FIG. Expertise on keywords extracted from user-written documents

To create a user profile. These profiles are periodically updated using recent documents created by users.

3 is a diagram of a generated dynamic user profile according to one embodiment of the present invention.

를 나타내는

의 평균을 계산하고 빈도 수를 고려하면 사용자의 동적 사용자 프로필을 생성한다. 첫 번째 단계에서는 동적 사용자 프로필을 생성하기 위해 사용자가 작성한 모든 문서에서 키워드를 추출하고 키워드별 전문성 점수

를 계산한다. 키워드별 정렬에서는 동일한 키워드에 대한

을 계산하기 위해 모든 키워드를 정렬하여 키워드 E, G와 같은 동일한 키워드를 찾아낸다. 세 번째 과정은 동일한 키워드의 상이한 전문성 점수의 평균을 계산하여 키워드 E의

는 (0.78 + 0.51)/2 = 0.645, 키워드 G의

는 (0.51 + 0.26)/2 = 0.385가 된다. 중복된 키워드의 존재는 사용자가 해당 키워드에 관심이 많고 자주 사용하는 키워드라는 것을 의미하므로 사용자 동적 사용자 프로필은 키워드,

, 빈도 수로 구성한다.
Referring to FIG. 3, when keywords A, B, C, D, E, F, and G extracted from a document created by the user (text 1, text 2, text 3)

Indicating

And calculates the dynamic user profile of the user in consideration of the frequency. The first step is to extract keywords from all the documents you have created to generate a dynamic user profile,

. Sorting by keyword is not

To find the same keywords such as keywords E and G by sorting all the keywords. The third step is to calculate the average of the different expertise scores of the same keyword,

(0.78 + 0.51) / 2 = 0.645, the keyword G

(0.51 + 0.26) / 2 = 0.385. The presence of duplicate keywords means that users are interested in and frequently use the keyword, so a user dynamic user profile can be a keyword,

, And frequency.

4 is a flowchart illustrating a dynamic user profile creation process according to an exemplary embodiment of the present invention.

Referring to FIG. 4, the server periodically detects recent activities of users and generates and modifies a dynamic user profile with recent activity information. Once a range of specific dates is entered (S1) to obtain current activity information, a list of all users subscribed to the social network is collected (S2). It is determined whether there are users satisfying the range of the specific date (S3). If there are users satisfying the range of the specific date in the collected list, The weights and preferences are calculated (S4 to S5). When the preference and the weight are calculated, the keyword of the m-th document created during the range of the specific date is extracted by the n-th user (S6). The preference degree and the weight of the document are applied to the extracted keyword, and the expertise score is calculated as in Equation (1) (S7). It is determined whether there is another document (S8). If there is another document, the process proceeds to step S5 to repeat the last document up to the last document (S9). If the expertise score for each keyword of the last document is calculated, Check. The same keyword is calculated as an average of expertise scores (S10). After all calculations are completed, a dynamic user profile is created and modified with the score and frequency of expertise by keyword (S11). It is determined whether there is another use (S12). If there is another user, the process proceeds to step S5 to repeat the process up to the last document (S13).

In social networks, the reliability of users is very important because many users share and distribute unverified information. Conventional techniques could not judge a malicious user because only the user profile information was analyzed or only the recent activity information was analyzed to determine whether the keywords were matched. In order to solve these problems, a reliability score is given to each user to perform a reliable expert search. There are currently no well-known social network services such as Facebook, Twitter, and LinkedIn that each user can rate each other. In the present invention, the users evaluate each other in the social network service. As a method for calculating the reliability score, we use the weighted average of ratings technique for the reliability among the users in the expert search.

와

사이에 신뢰성을 계산한 식이다. 이때,

는

와

사이에 있는 모든 소셜 네트워크 경로에 있는 사용자들 사이에 신뢰성의 합,

는

와 직접적으로 연결된 사용자만의 신뢰성에 가중치를 부여한 값이다.
Reliability between two users uses reputation scores of users in the network path of two users connected by a social network. For example, a friend who is kind to a particular user may be an unfriendly person to another. Even if I rate a friend user with a high score, another user can give a different rating, so I need a way to evaluate the user more objectively. Therefore, through the network of users connected to the social network, other users evaluate a specific user to give a reputation score, and calculate the reliability by summing each score. (Equation 4)

Wow

The reliability is calculated. At this time,

The

Wow

The sum of credibility among users in all social network paths between,

The

And the reliability of only the user directly connected with the user is weighted.

--- (Equation 4)

FIG. 5 is a diagram for creating a trust network among users according to an exemplary embodiment of the present invention.

는 0.6 + 0.833 + 0.733 + 0.525 = 2.691,

는 (0.9 X 0.6) + (0.9 X 0.833) + (0.5 X 0.733) + (0.5 X 0.525) = 1.9187,

는 1.9187/2.691 = 0.713가 된다.
Referring to FIG. 5, the users A and B are friends B and C, and D, F and E are users B and C, respectively. To obtain the reliability for users A and F, the path between users A, B and C and F is directly obtained. Path 1 is ABF, path 2 is ABDF, path 3 is ACDF, and path 4 is ACDBF. (0.9 + 0.7 + 0.9) / 3 = 0.833, (0.5 + 0.8 + 0.3) / 3 = 0.733, (0.5 + 0.8 + 0.5 + 0.3) / 4 = 0.525. Therefore, using the equation (4) for reliability with F considering all paths of A

0.6 + 0.833 + 0.733 + 0.525 = 2.691,

(0.9 X 0.6) + (0.9 X 0.833) + (0.5 X 0.733) + (0.5 X 0.525) = 1.9187,

Is 1.9187 / 2.691 = 0.713.

6 is a flowchart illustrating a reliability evaluation process of a social relationship according to an embodiment of the present invention.

Referring to FIG. 6, upon receipt of two users i and j to evaluate reliability (S21), a list of all users existing between the two users i and j is collected (S22). Users who do not have a reputation score are set to "0" (S24), and then users i and j (S25). If there are other users in the middle of the distance between the user i and j, the calculation is performed as shown in (Equation 4). First, users directly connected to the user i are selected (S26), and all paths between the selected users and the user j are collected (S27). The average of the reputation scores of the users in each path is calculated (S28), the weight is given to each path calculated as (Equation 4), and divided by the sum of average reputation scores of all paths, (S29). On the other hand, if the users i and j are directly connected and there is no other user in the middle, the reputation score between them is used as reliability (S30 to S31).

의 프로필

과 사용자

가 검색한 질의 클러스터

의 유사도는 (식 5)과 같다. 이때,

는 질의 클러스터와 일치하는 키워드의 수,

는 질의 클러스터와 일치하는 키워드의

,

는 키워드가 발생한 최대 수,

는 키워드의 빈도수,

는 사용자

에 대한 사용자

의 신뢰성이다. 소셜 네트워크에 연결된 사용자들에 대해 1촌 관계부터

을 계산하여 임계치 이상의 유사도를 갖는 사용자들을 검색한다. 검색할 범위는 6촌 관계까지이다. 하지만 6촌 관계는 거의 전 세계 사람을 고려하게 되어 계산 비용이 상당하기 때문에 3촌 관계까지로 한다. 검색된 결과는 유사도에 따라 정렬되어 사용자에게 전달된다. 따라서 확장된 질의 클러스터와 동적 사용자 프로필을 비교하여 정확도가 높은 결과가 도출되며 평판 점수를 활용하여 신뢰성을 보장된다.
In order to search for an expert in a social network, a similarity is determined by comparing a profile of a connected user with a search query desired by a user in a social network relationship. When searching for experts using only the queries requested by the user, the accuracy decreases. Therefore, in the present invention, all the synonyms, dialects, and the like are searched using the ontology for the query term. Connected to social networks

Profile of

And users

Query cluster

Is the same as (Equation 5). At this time,

The number of keywords matching the query cluster,

Of the keywords matching the query cluster

,

Is the maximum number of occurrences of the keyword,

The frequency of the keyword,

User

User for

. From a one-village relationship for users connected to social networks

And searches for users having a degree of similarity equal to or greater than the threshold value. The search range is up to 6 villages. However, since the relationship between the six villages is considered almost entirely of people in the world, the cost of the calculation is considerable, so that the relationship between the three villages is made. The retrieved results are sorted according to the degree of similarity and transmitted to the user. Therefore, the extended query cluster is compared with the dynamic user profile, resulting in highly accurate results, and reputation scores are utilized to ensure reliability.

--- (식 5)

--- (Equation 5)

FIG. 7 is a diagram comparing the similarity between a query and a user profile according to an embodiment of the present invention.

= 0.6, 질의 클러스터는 키워드 E와 D, 동의어 EA와 DA, 그리고 방언 EB와 DB로 이루어져 있다고 가정한다. 도 3과 같은 사용자 F에 대한 동적 사용자 프로필이 생성되어 있다면, 질의 클러스터와 동적 사용자 프로필을 비교했을 때 키워드 E와 D가 일치하게 된다. 빈도 수와 일치한 키워드들의 수를 고려한 평균 전문성 점수

= [{0.645 X (2/2)} + {0.78 X (1/2)}]/2 = 0.5175를 계산하고 도 5에 나타난 신뢰성 점수

= 0.713을 합하여 사용자 F에 대한 유사도

= (0.6 X 0.5175) + (0.4 X 0.713) = 0.5957이 계산된다. 질의 클러스터와 일치하는 사용자 F, H, J, K에 대한 유사도를 반복적으로 계산하고 유사도가 높은 순서대로 결과가 출력한다.
Referring to FIG. 7, in a first step, when a user requests a query, a cluster for a query is formed and a search starts from a user in a one-village relationship. To measure similarity

= 0.6, and the query cluster is assumed to consist of keywords E and D, synonyms EA and DA, and dialects EB and DB. If a dynamic user profile is created for user F as shown in FIG. 3, keywords E and D match when a query cluster is compared with a dynamic user profile. Average expertise score, taking into account the number of keywords matching the frequency

= [{0.645 X (2/2)} + {0.78 X (1/2)}] / 2 = 0.5175,

= 0.713 to calculate the similarity to user F

= (0.6 X 0.5175) + (0.4 X 0.713) = 0.5957 is calculated. The similarities of the users F, H, J, and K that match the query cluster are calculated repeatedly, and the results are output in order of high similarity.

FIG. 8 is a flowchart illustrating an expert search process according to an exemplary embodiment of the present invention.

Referring to FIG. 8, when the user i inputs a search query Q and a maximum ranking K (S41), the server determines whether there is a synonym, a thesaurus, a dialect, etc. in Q (S42). If there is Q, the search query cluster QC expanded with Q, synonyms, synonyms, and dialects is generated (S43). Otherwise, the unexpanded Q is used as it is (S44). Since it is inefficient to consider all the users of the social network, the search starts from the one-village user of the user i (S45 to S46). A list of the users of the one-person relationship of the user i is collected (S47), and a dynamic user profile is compared between the QC and one-dimensional users as shown in (Equation 5). The degree of similarity is calculated by adding a reliability score to the compared value (S48). When the similarity degree is calculated for all the users in the one village relationship, the similarity degree list is generated with the users having the greatest similarity as the first ranking (S49). It is determined whether the last rank of the similarity satisfies K (S50). If the last rank of the similarity satisfies K, all experts are ranked and output as a result. On the other hand, if the last rank of the degree of similarity does not satisfy K, the degree of similarity is calculated from the users of the two-village relationship having the following relationship (S51). Since the number of users is more than 4 villages, the search efficiency is lowered. Therefore, the degree of similarity is measured only up to 3 villages.

While the present invention has been particularly shown and described with reference to exemplary embodiments thereof, it is to be understood that the invention is not limited to the disclosed exemplary embodiments, but, on the contrary, is intended to cover various modifications and equivalent arrangements included within the spirit and scope of the invention.

1: Social information collecting section
2: Social Information Analysis Department
3: Dynamic User Profile Creation Unit

Claims (13)

A dynamic user profile manager for generating and updating a dynamic user profile by calculating a professionalism score based on the number of documents created on the date range, the preference for the document, and the weight assigned according to the number of participants;
A social network reliability calculation unit for calculating a reliability score of a social network for a user with a reputation score assigned to a network path between users connected to the social network; And
And an expert search unit for comparing the dynamic user profiles corresponding to the search query, adding the reliability scores to the compared values, and searching for experts according to the similarity rank,
The social network reliability calculation unit collects network paths among users, selects users having a specific user and a network path among users, collects all network paths existing between the specific user and selected users A dynamic user profile calculating a reliability score of the specific user by dividing the weight value assigned to each network path by the sum of average reputation scores of all paths, calculating a mean value of reputation scores existing in all the network paths, Expert Search System Using.
The method according to claim 1,
The dynamic user profile management unit,
A social information collection unit for collecting the document, scrap, recommendation and comment on the document, and social information including the number of participants;
A social information analyzer for extracting keywords by using a text mining technique in the document, calculating the preferences by analyzing scraps, recommendations, and comments on the documents, and assigning weights according to the number of participants; And
A dynamic user profile generating unit that calculates the expertise score with the preference for the keyword and the expertise score with the keyword, and generates a new dynamic user profile with keyword specificity score and keyword frequency, .
(a) generating and updating a dynamic user profile by calculating a professionalism score with a weight given according to the number of documents created on a range set date, the preference for the document, and the number of participants;
(b) calculating, by the social network reliability calculation unit, a reliability score of the social network for the user with a reputation score assigned to a network path between users connected to the social network; And
(c) comparing the dynamic user profile corresponding to the search query in response to the search query, adding the reliability score to the compared value, and searching for the expert by the order of similarity,
The step (b)
Collecting network paths between users;
Selecting a specific user and users having a network path among the users;
Collecting all network paths existing between the specific user and selected users;
Calculating an average value of reputation scores in all the network paths; And
Dividing the weight assigned to each network path by the sum of average reputation scores of all paths, and calculating a reliability score of the specific user.
The method of claim 3,
The step (a)
Collecting the document, scrap, recommendation and comment on the document, and social information including the number of participants;
Extracting keywords from the document using a text mining technique, calculating the preferences by analyzing scraps, recommendations, and comments on the documents, and assigning weights according to the number of participants; And
Calculating a preference score for the keyword and the expertise score using the weight, and generating a new dynamic user profile based on the keyword expertise score and keyword frequency.
5. The method of claim 4,
The document (

) ≪ / RTI >

) Is the expert search method using the dynamic user profile and social network reliability calculated by (Equation 1) below.

--- (1)
here,

Document

Weight,

Document

Preference for,

The number of documents created by the user,

Is a constant.
6. The method of claim 5,
The document (

) ≪ / RTI >

) Is the expert search method using the dynamic user profile and social network reliability calculated by (Equation 2) below.

--- (Equation 2)
here,

Document

The number of users who participated in,

Is a constant.
6. The method of claim 5,
The document (

) Preference for

) Is the expert search method using the dynamic user profile and social network reliability calculated by (Equation 3) below.

--- (Equation 3)
here,

Is a constant,

Document (

),

Document (

),

The

, Which is a rating of 1, 0, and 0.5, depending on whether they are positive, negative, or neutral.
delete The method of claim 3,
The reliability score of the particular user (

) Is the expert search method using the dynamic user profile and social network reliability calculated by (Equation 4) below.

--- (Equation 4)
here,

The

Wow

The sum of credibility among users in all social network paths between,

The

And the reliability of only the user directly connected with the user is weighted.
The method of claim 3,
Expert search method using dynamic user profile and social network reliability that calculates reputation score as reliability score when there is no other user among users.
The method of claim 3,
The step (c)
Generating a search query cluster extended with synonyms, synonyms, and dialects using the ontology in response to the search query;
Collecting a list of users directly connected to the search query user;
Comparing the search query cluster with a dynamic user profile of the directly connected users;
Calculating a degree of similarity by adding a reliability score to the compared value;
Generating a similarity score list in descending order of the degree of similarity; And
Determining a ranking of all experts above a threshold value and outputting the results as a result if the last rank of the similarity is greater than or equal to a threshold value and continuously calculating the similarities of users indirectly connected to the user if the last rank of the similarity rank is below a threshold, Expert search using user profile and social network credibility.
12. The method of claim 11,
A dynamic user profile using the search query as it is when there is no synonym, synonym, or dialect; and a professional search method using the reliability of a social network.
12. The method of claim 11,
The user connected to the social network

Profile of

And users

Query cluster

Similarity (

) Is the expert search method using the dynamic user profile and social network reliability calculated by (Equation 5) below.

--- (Equation 5)
here,

The number of keywords matching the query cluster,

Of the keywords matching the query cluster

,

Is the maximum number of occurrences of the keyword,

The frequency of the keyword,

User

User for

.

Images