KR102106203B1 - Expert recommending method and system for providing social network system based question and answer service - Google Patents

Abstract

Disclosed is an expert recommendation method and an expert recommendation system for providing a question and answer service based on a social network system. An expert recommendation method according to an embodiment of the present invention comprises: extracting a query keyword from the query as a query is requested from a first user terminal using a social network system, and among user terminals using the social network system , Searching for a second user terminal that has registered a field of interest associated with the query keyword, configuring the second user terminal into a group of experts who can respond to the query, and including the group of experts, requesting the query And outputting the expert recommendation result according to the first user terminal.

Description

An expert recommendation method and an expert recommendation system for providing a question and answer service based on a social network system {EXPERT RECOMMENDING METHOD AND SYSTEM FOR PROVIDING SOCIAL NETWORK SYSTEM BASED QUESTION AND ANSWER SERVICE}

The present invention relates to an expert recommendation technology for providing a question and answer service based on a social network system (SNS), by recommending an expert terminal capable of quickly and accurately responding to a user query in consideration of an interest field of a user terminal and a human network and response quality, It is related to the expert recommendation method and expert recommendation system that can increase the utilization and user satisfaction of the Q & A service through SNS.

Social network system (SNS) refers to an online platform that can share information created by users with other users who have a personal relationship, and typically, SNS such as Twitter, Facebook, Instagram, etc. Is being used by

In addition, in social network systems such as Quora and LinkedIn, in addition to communication and information sharing, a query response service is provided that provides a response written by an acquaintance on an SNS to a query created by a user.

However, the existing question-and-answer service through SNS is difficult to provide high-quality answers by experts because the personal relationship of the user who requested the query is limited, and includes various information including malicious information, unreliable information, and unnecessary information. There is a limit to filtering and providing high-quality answers that match the queries requested by users because vast information exists without limitations on the SNS.

For this reason, an expert recommendation service has been proposed that recommends experts who can provide high-quality and reliable answers to user inquiries in connection with the existing question-and-answer service through SNS. For example, social network users directly input Techniques for recommending based on one's own profile and techniques for analyzing and recommending each user's social behavior are used.

On the other hand, in the case of a technique of recommending a user with a high degree of similarity between a profile and a user query as an expert, the accuracy of the expert recommendation may be lowered if the user does not update the profile and creates an old or false profile.

In addition, in the case of a technique of recommending an expert by analyzing a user's social behavior, it is difficult to recommend an accurate expert to a user with little personal relationship because the recommendation is made within the personal relationship between the querying user.

In addition, the existing expert recommendation technique recommends the expert without considering the quality of the answer written by the expert, so it can be difficult to determine whether the user can actually receive a high-quality response from the expert quickly.

Accordingly, by analyzing the user's social behavior in the social network system, each user's interests, personal relationships, and response quality are determined, and experts with rich personal relationships and good response quality are recommended for each interest field related to the user query. There is a need.

According to an embodiment of the present invention, considering user interests, personal relationships and response quality of user terminals using a social network system (SNS), while making user queries their latest interests, other user terminals in the interest field By selecting and recommending a user terminal that is rich in personal relationship and has good response quality as an expert terminal to request a question and answer, so as to improve the accuracy and reliability of expert recommendation and maintain high satisfaction with the question and answer service through SNS It aims to do.

In addition, an embodiment of the present invention determines a recent interest through analysis of a user's social behavior, and selects and recommends an expert for answering a question from among users identified as an interest related to the user's query, so that an existing profile-based expert In the recommendation technique, if the profile was created a long time ago or was made false, the goal is to solve the problem that the accuracy of expert recommendation is lowered and to maintain the latestness of each user's interest.

In addition, the embodiment of the present invention rebuilds the human network on the social network system for each interest field and analyzes the degree of having a human relationship between users of the same interest field, so that an expert who has a high influence within the interest field is expert By selecting as and recommending, users with high professionalism in querying when there are few users who have a personal relationship in the technique of recommending experts within a human network centering on existing querying users or when their interests are not diverse It aims to solve the problem difficult to recommend.

In addition, the embodiment of the present invention evaluates the response quality by analyzing at least one of the positive response (comment, sympathy, sharing, etc.) of the other user terminal to the past response of the user terminal and the response speed and similarity between the past response and the field of interest. By doing so, it is an object to recommend a user who can answer other user's queries quickly and accurately based on the response quality.

In addition, an embodiment of the present invention is to increase the accuracy and reliability of expert recommendation by analyzing a user's query and matching a group of experts using a hierarchical structure of words.

An expert recommendation method for providing a query response service based on a social network system according to an embodiment of the present invention comprises: extracting a query keyword from the query as a query is requested from a first user terminal using the social network system; , Among the user terminals using the social network system, searching for a second user terminal that has registered an interest field associated with the query keyword, and configuring the second user terminal into a group of experts who can answer the query, And outputting the expert recommendation result according to the query request to the first user terminal, including the expert group.

In addition, the expert recommendation system for providing a query response service based on a social network system according to an embodiment of the present invention extracts a query keyword from the query as a query is requested from a first user terminal using the social network system Keyword extraction unit, among the user terminals using the social network system, a terminal search unit for searching for a second user terminal that has registered a field of interest associated with the query keyword, and an expert capable of answering the second user terminal to the query A group configuration unit configured as a group, and a result output unit including the expert group and outputting expert recommendation results according to the query request to the first user terminal.

According to an embodiment of the present invention, by selecting an expert to answer a query in consideration of both the latest interests in social users' interests and the relationship between users and the quality and speed of past responses, the accuracy and reliability of expert recommendations are improved and SNS is improved. Through this, it is possible to maintain high satisfaction with the Q & A service.

In addition, according to an embodiment of the present invention, by analyzing social behavior data of users using SNS, the latest interests of each user are determined, and the human networks between SNS users are re-established by interests, and user queries and By selecting users with rich interests in the relevant fields and having relevant interests as experts to request questions and answers, it is possible to keep users' interests up to date and recommend experts with high influence in the relevant fields.

In addition, according to an embodiment of the present invention, by evaluating a response quality through analysis of a user's past response, a user who can quickly and accurately answer another user's query can be recommended as an expert.

In addition, according to an embodiment of the present invention, by using a hierarchical structure of words to recommend a group of experts including experts matching sub keywords of the keyword as well as keywords in the user query, the accuracy and reliability of expert recommendation Can be increased.

In addition, according to an embodiment of the present invention, it can be used as a question and answer system that can hear expert opinions on products or research.

1 is a block diagram showing the configuration of an expert recommendation system according to an embodiment of the present invention.
2 is a block diagram showing the configuration of an expert recommendation system according to another embodiment of the present invention.
3 is a block diagram showing a detailed configuration of an interest index calculation unit in an expert recommendation system according to another embodiment of the present invention.
4 is a block diagram showing a detailed configuration of a personal relationship index calculator in an expert recommendation system according to another embodiment of the present invention.
5 is a block diagram showing a detailed configuration of a response quality index calculation unit in an expert recommendation system according to another embodiment of the present invention.
6 is a diagram illustrating an example of rebuilding a human network between user terminals according to interests in an expert recommendation system according to embodiments of the present invention.
7 is a diagram illustrating an example of providing a recommendation result to a group of experts matching a user query in the expert recommendation system according to embodiments of the present invention.
8 is a flowchart illustrating a procedure of a method for recommending experts according to an embodiment of the present invention.

Hereinafter, an expert recommendation method and system for providing a question and answer service based on a social network system according to an embodiment of the present invention will be described in detail with reference to the accompanying drawings. However, the present invention is not limited or limited by the embodiments. The same reference numerals in each drawing denote the same members.

1 is a block diagram showing the configuration of an expert recommendation system according to an embodiment of the present invention.

Referring to FIG. 1, the expert recommendation system 100 according to an embodiment of the present invention includes a keyword extraction unit 110, a terminal search unit 120, a group configuration unit 130, and a result output unit 140. It can be configured to include. In addition, according to an embodiment, the expert recommendation system 100 adds a collection unit 150, a field of interest registration unit 160, a network reconstruction unit 170, a response quality evaluation unit 180, and a database 190, respectively. Can be configured.

The keyword extraction unit 110 extracts a query keyword from the query as a query is requested from a first user terminal using a social network system.

For example, the keyword extracting unit 110 may analyze the above query and extract a query keyword ('figure') when a query is requested from the user terminal through the SNS ('When will the One Piece 6 rupee figure be released?'). have.

The terminal search unit 120 searches for a second user terminal that has registered a field of interest associated with the query keyword from among user terminals using a social network system.

For example, the terminal search unit 120 may search the database 190 for a second user terminal that has registered a query keyword (“Luffy Figure”) as a field of interest. If the second user terminal that registers the same keyword as the query keyword ('figure') as the field of interest is not in the database 190, the keyword ('goods') of the query keyword (' figure ') or another query keyword in the query (' One Piece ') as a field of interest, the second user terminal can be searched in the database 190.

At this time, the terminal search unit 120 may check a topic according to the context of the query, and search for a second user terminal that registers an interest field associated with the query keyword for each topic.

For example, the terminal search unit 120 searches for a second user terminal that registers a field of interest related to a query keyword ('figure') as a subject ('ani one piece') according to the context of the query to obtain accuracy of expert recommendation. It can be further increased.

According to an embodiment, the expert recommendation system 100 further includes a collection unit 150, an interest registration unit 160, a network reconstruction unit 170, a response quality evaluation unit 180, and a database 190, respectively. You can.

The collection unit 150 is a social action of each user terminal recorded by including a document in the form of text of at least one of posts and comments created or shared by each user terminal during a predetermined period of time based on when the query is requested. Data is collected from the social network system.

Here, the social action may refer to all actions performed by the user terminal on the social network system. For example, write a post / comment, share a written post / comment, or respond positively or query like “Like!” It may be responding to or having a personal relationship with another user terminal.

That is, the collection unit 150 includes at least one of the posts (documents) written by each of the user terminals during the period and the posts shared with the comments, the number of times shared, the number of responses, the number of comments, the date of creation, and the history of query responses. Social behavior data may be collected.

Also, the collection unit 150 may collect a human network indicating a connection relationship between user terminals on the social network system.

The interest register 160 analyzes the social behavior data collected from the social network system, and registers the interest in each of the user terminals.

For example, the interest registration unit 160 assigns a keyword score according to a term frequency-inverse document frequency (TF-IDF) technique for each keyword of interest extracted from the document, and each document in which the keyword of interest is extracted With respect to, a reliability score is given in consideration of the number of times the reaction is input and the date and time of document creation, and a certain number of interest keywords are selected in the order of the keyword score and the reliability score, and the selected interest keywords are Each user terminal can be registered as a field of interest.

Here, the TF-IDF technique is a technique for calculating a statistical number indicating how important a word is in a specific document among a group of documents composed of several documents, and the TF value is a word frequency and a value indicating a frequency in which a specific word appears in a document The IDF value represents a value indicating how many documents contain a specific word in the entire document at the frequency of reverse documents, and the keyword detection unit 312 calculates the keyword score using the TF-IDF value multiplied by the TF value and the IDF value. Can be given.

For example, if there are 100 documents (written or shared) that user terminal A has active, 8 of which contain the word 'Hadoop', and the word 'Hadoop' appears in document B of one of them 5 times. , The keyword detector 312 may calculate the TF-IDF value as '12 .6286 'by setting the TF value for document B to' 5 'and the IDF value to' 2.52573 '.

The higher the TF-IDF value, the keyword detection unit 312 may determine that it is an important word and extract the word as a keyword of interest.

At this time, the collection unit 150, if a certain time has elapsed from the time when the query is requested, based on the elapsed time, the social behavior data of each user terminal recorded during the previous period, the social network system It can be re-collected from, and the interest registration unit 160 may update the registered interest by analyzing the re-collected social behavior data.

Accordingly, the interest registration unit 160 may maintain the latestness of the interest field of each user terminal, thereby solving the problem of less accuracy when recommending an expert based on an old profile that the user directly registered.

The network reconstruction unit 170 re-establishes a human network built between user terminals on a social network system according to interests.

For example, referring to FIG. 6, the network reconstruction unit 170 divides a human network between SNS user terminals that have registered various interests as shown in FIG. 6 into interests A and B, respectively. Thus, it is possible to rebuild the human network between terminals having the same interest field A shown in FIG. 6 (ii) and the human network between terminals having the same interest field B shown in FIG. 6 (i). have.

The terminal search unit 120 searches for the second user terminal in the re-established human network regarding the field of interest associated with the query keyword, so that the personal relationship between the first user terminal requesting the query is narrow or an acquaintance with various interests Even if there is no, it is possible to easily search for a second user terminal whose user query is a specialized field in a human network configured for each field of interest.

In addition, the terminal search unit 120 may search for a user terminal having a set number of connected terminals or more in the rebuilt human network as the second user terminal.

Specifically, the terminal search unit 120 calculates the human relationship index of the user terminal using the PageRank algorithm based on the reconstructed human network as shown in (ii) of FIG. 6 when the query keyword is related to the interest field A. You can.

The terminal search unit 120 counts the number of connected human terminals, that is, the number of connected terminals, by applying the PageRank algorithm that calculates the score per web page to the human relationship through the link relationship between the web pages. Then, a PageRank score can be assigned according to the number of counted terminals.

The higher the number of connected terminals (the higher the PageRank score) is, the higher the number of connected terminals is, the higher the personality index is, and by searching for the second user terminal that ranks higher according to the personality index, the influence is greater in the field. It is possible to make recommendations from experts with excellent expertise.

In this case, the terminal search unit 120 may search for the second user terminal by giving priority to the user terminal connected to the first user terminal in the rebuilt human network.

For example, if the re-established human network as shown in (ii) of FIG. 6 includes an acquaintance who has a personal relationship with the first user terminal that requested the query, the corresponding person is searched as the second user terminal, and the queryer himself You can easily inquire users who are related to and increase the intimacy with acquaintances.

The response quality evaluation unit 180 evaluates the response quality for a past response submitted by the user terminal in the case of a user terminal that includes a query response history in the social behavior data.

For example, the response quality evaluator 180 may numerically evaluate the response quality by considering at least one of the similarity between the past response and the field of interest, the response time, and the response of the other user terminal to the past response. .

The terminal search unit 120 may search the user network as the second user terminal in the order of the numerical value indicating the response quality in the rebuilt human network.

That is, the terminal search unit 120 may search for the user terminal having the highest response quality index as the second user terminal, so that the quality of the past response is considered when recommending experts.

The group configuration unit 130 configures the second user terminal as a group of experts who can answer the query.

As an example, the group configuration unit 130 has a high interest index among user terminals that have registered interest fields related to a query keyword, and a high personal relationship index in a human network reconstructed by interest fields, and a high response quality index. The second user terminal can be configured as an expert group.

In addition, the group configuration unit 130 is a second user terminal having a high interest index among user terminals that have registered interest fields related to a query keyword, and a second user having a high personal relationship index in a human network reconstructed for each interest field. The expert group may be configured to include both a terminal and a second user terminal having a higher response quality index.

In addition, the group configuration unit 130 may match the query group, and maintain it in the database 190 by matching the query group, thereby facilitating the search of the expert group through subsequent query keyword matching.

Also, the group construction unit 130 may match a group of sub-experts composed of each sub-keyword for the query keyword to the sub-keywords.

For example, the group configuration unit 130 includes a sub-expert group consisting of sub-keywords ('Starbucks', 'coffee bean', 'twosome', etc.) for the query keyword ('brand coffee'), respectively. You can match sub keywords.

In addition, the group configuration unit 130 generates a tree (see 710 in FIG. 7) in which the expert group and the sub-expert group are arranged as hierarchical structures of matched keywords as nodes, and maintained in the database 190. You can.

In addition, the database 190 maintains in the database 190 for each of the user terminals, analysis data of at least one of social behavior data collected from SNS, interests, and re-established human networks and response quality regarding the interests. You can.

The result output unit 140 includes the expert group and outputs the expert recommendation result according to the query request to the first user terminal.

In addition, the result output unit 140 may output expert recommendation results, including a group of experts matching a keyword matching the query keyword and a group of sub-experts diverging from the expert group.

Specifically, when the same query keyword as the query is extracted from the new query, the result output unit 140 selects the expert group that matches the keyword matching the query keyword in the tree maintained in the database 190. The search may search for a sub-expert group diverging from the expert group, and output the expert recommendation result including the expert group and the sub-expert group.

In other words, the result output unit 140 compares and matches the query keyword with the tree, searches for a group of experts matching the keyword matching the query keyword, and considers the hierarchical structure of words between the matched keywords, In the tree, sub-expert groups that branch from the expert group can be searched.

The result output unit 140 may select some user terminals according to the number of branches diverging from the expert group among all user terminals belonging to the expert group and user terminals belonging to the lower expert group, and include the result in the expert recommendation result.

For example, referring to FIG. 7, when the query keyword extracted from the user query matches the expert group B, the result output unit 140 is the expert group B and the expert group C matching the subkeyword of the query keyword It is possible to output expert recommendation results including, D, and E.

At this time, the result output unit 140 in the expert group C, D, E each of the user's terminal according to the number of each sub-keyword ('3') in the order of the expert index ('1/3') Only (Top Expert) can be selected and included in expert recommendation results.

As described above, according to an embodiment of the present invention, the accuracy and reliability of expert recommendation are selected by selecting an expert to answer a query in consideration of both the latest interests of social users and the relationship between users and the quality and speed of past responses. It can improve and maintain high satisfaction with Q & A service through SNS.

2 is a block diagram showing the configuration of an expert recommendation system according to another embodiment of the present invention.

Referring to FIG. 2, the expert recommendation system 200 according to another embodiment of the present invention may include a user expertise analysis unit 210 and a user query analysis unit 220.

The user expertise analysis unit 210 functions to analyze the expertise of each user terminal based on the SNS activity history of each of the user terminals collected from the social network system.

Here, the professionalism of the user terminal means the degree of information rich in the field of interest of the user terminal, and the user expertise analysis unit 210 may analyze the expertise of the user terminal by an expert index for each field.

To this end, the user expertise analysis unit 210 may analyze the data collected from the social network system to determine a field of interest and calculate an expert index in the field of interest.

The user expertise analysis unit 210 shares the posts or comments, the date and time of writing, and each user terminal shared by each user terminal for a predetermined period of time (for example, 'last month') in order to analyze a field rich in SNS activity history. Data on at least one social action among text, number of sharing, number of empathy, number of positive responses, and question and answer history may be collected.

Here, the creation date and time may be collected to analyze a field of recent interest based on a post or comment recently created by the user terminal, and a query response history may be collected to analyze the quality of the response created during the corresponding period.

In addition, the user expertise analysis unit 210 may further collect a human network indicating a connection relationship between user terminals on the social network system.

As an example, as shown in FIG. 2, the user expertise analysis unit 210 includes an interest index calculation unit 211, a personal relationship index calculation unit 212, a response quality index calculation unit 213, and an expert index calculation unit ( 214).

The interest index calculation unit 211 analyzes the collected SNS activity history, extracts keywords of interest, and calculates the interest index based on the number of social actions and the latestness of each keyword of interest.

Also, the interest index calculation unit 211 may register at least one keyword of interest in which the interest index is high, as an interest field (specialized field) of each user terminal.

For example, the interest index calculation unit 211 extracts 'automobile fair', 'Japan travel', and 'figures' as keywords of interest from posts and comments created or shared by the user terminal in the SNS activity history, posts By assigning a score according to the number of times a comment is made or shared, an interest index is calculated for each keyword of interest, and the interest keyword 'figure' having a high interest index is determined as an interest of the user terminal. .

In particular, the interest index calculation unit 211 may determine an interest field based on a recent interest of the user terminal by calculating a interest index by assigning weights according to whether posts and comments are recently written.

The human relationship index calculating unit 212 analyzes the collected human network and calculates the human relationship index of the user terminal for each interest.

As an example, the human relationship index calculating unit 212 rebuilds the human networks between user terminals using the social network system by interests, and rebuilds the human networks for each of the user terminals in the human network for each interest field. A score can be calculated according to the number of times a relationship is established to calculate a personal relationship index.

Through this, it is possible to discriminate the user terminal having high influence in the corresponding field according to the degree of establishing a human relationship between the user terminals connected to the same interest field.

The response quality index calculating unit 213 calculates a response quality index based on the response speed and the reliability of the response to the past response of each user terminal.

For example, the response quality index calculating unit 213 analyzes at least one of the similarity of the response and the field of interest, the response time, the response speed, the satisfaction of the queried user, and the response (eg, comments, sharing) of other user terminals. Calculate the response quality index.

Accordingly, the response quality index calculating unit 213 has a relatively short response time, has a high response speed, has a high degree of similarity with the field of interest, and is relatively high with respect to a user terminal having a positive past response from another user terminal. Response quality index can be calculated.

The expert index calculation unit 214 calculates an expert index by adding an interest field index, a personal relationship index, and a response quality index.

The expert index calculating unit 214 may analyze the expertise in the field of interest determined according to the interest field index based on the expert index.

For example, the expert index calculating unit 214 is an expert according to the interest index '34' + the personal relationship index '65' + the response quality index '0' when the interest field of the user terminal A is determined as 'figure' The index '99' can be calculated.

In addition, the expert index calculating unit 214 determines the expert index '104' by the interest field index '24' + the personal relationship index '50' + the response quality index '30' for the user terminal B determined to be the same interest field. Can be calculated.

Accordingly, the expert index calculating unit 214 may analyze that the user terminal B having a high expert index is high in 'figure' in the corresponding field.

The expert index calculator 214 may derive an expert index according to Equation 1 by synthesizing the user's interest index, personal relationship index, and response quality index.

In Equation 1, keyword is an index of interest, Relations is a personal relationship index, and QAR is a response quality index. The sum of the weights α, β, and γ is expressed by 1.

The expert index calculating unit 214 may configure user terminals into expert groups for each field in consideration of interests, personal relationships, and response quality using the expert indexes.

The expert index calculating unit 214 may configure a group of experts based on the ontology constructed based on the hierarchical relationship of words. That is, the expert index calculator 214 may include the user as a member in the ontology group by using the expert index determined for each user.

The user query analysis unit 220 functions to recommend a group of experts who can respond to a user query based on the expert index when a query is requested from the user.

As an example, as illustrated in FIG. 2, the user query analysis unit 220 may include a query keyword analysis unit 221, a query-expert field matching unit 222, and an expert group recommendation unit 223. .

The query keyword analysis unit 221 extracts a query keyword by analyzing the query when a query is requested from the user terminal through SNS.

The query-expert field matching unit 222 searches for a group of experts who have the query as a field of interest by comparing each with a keyword matched to a group of experts who have previously constructed a query keyword. Here, the keyword matched to the expert group may mean an interest field of the user terminal belonging to the expert group.

Also, the query-expert field matching unit 222 may search for a group of sub-experts matched to the sub-keyword of the query keyword according to the hierarchical structure of words.

The query-expert field matching unit 222 may search for a group of experts matched with the query based on Equation (2). EG where _i is a group of experts in the expert group i, _i SEG is a group of the sub-control branch from EG _i, SEG _max is the total number of sub-control groups branched in the EG _i.

The expert group recommendation unit 223 recommends the expert group and the sub-expert group to the user terminal requesting the query according to the search.

At this time, the expert group recommendation unit 223 may select and recommend some user terminals in consideration of the number of sub-expert groups when recommending user terminals belonging to the sub-expert group.

In other words, the expert group recommendation unit 223 is an expert group capable of responding to a user query by including not only an expert group having a query keyword included in the user query as a field of interest, but also an expert group for sub-keywords of the query keyword in the recommendation result. The accuracy of election can be improved.

For example, referring to FIG. 7, when the query keyword extracted from the user query matches the expert group B, the expert group recommendation unit 223 is the expert group B and the expert group matched with the sub-keyword of the query keyword. Expert recommendation results can be output, including C, D, and E.

At this time, the expert group recommendation unit 223 includes some users corresponding to the top ('1/3') in the order of expert index according to the number of each sub-keyword ('3') in each of expert groups C, D, and E. The terminal (higher expert) can be selected and included in the expert recommendation result.

As described above, according to the present invention, each user's interests, personal relationships, and response quality are determined through analysis of the user's social behavior in the social network system, and the human relationship is rich in the interests associated with the user query and the response quality is good. By recommending the user as an expert to request a question and answer, the accuracy and reliability of the expert recommendation can be improved and high satisfaction with the question and answer service through SNS can be maintained.

3 is a block diagram showing a detailed configuration of an interest index calculation unit in an expert recommendation system according to another embodiment of the present invention.

Referring to FIG. 3, the interest index calculation unit 211 may include a keyword analysis unit 310 of interest, a document reliability calculation unit 320, and an index calculation unit 330.

The interest keyword analysis unit 310 collects the SNS activity history on the user's social network system and extracts the keyword through the morpheme analyzer.

For example, the keyword analysis unit 310 of interest may include a morpheme analysis unit 311 and a keyword detection unit 312.

The morpheme analysis unit 311 extracts a noun using a morpheme analyzer from a text form document such as a text or shared data written by a user terminal on a social network, and then removes stopwords. Here, the morpheme analyzer can be exemplified by the 'Hannam morpheme analyzer' developed by the Korea Advanced Institute of Science and Technology (KAIST).

The keyword detection unit 312 applies TF-IDF techniques to words obtained from the morpheme analysis unit 311 to extract important words frequently used by users on social networks as keywords of interest.

The TF-IDF technique is a technique for calculating a statistical value indicating how important a word is within a specific document among a group of documents composed of several documents, and can be used to extract a key word of the document.

Here, the TF value is a word frequency, which is a value indicating the frequency with which a specific word appears in a document, and the IDF value is a reverse document frequency, and may indicate a value indicating how many documents contain a specific word in the entire document.

For example, if there are 100 documents (written or shared) by user terminal A, 8 of which contain the word 'Hadoop', and the word 'Hadoop' appears in document B of one of them 5 times. , The keyword detector 312 may calculate the TF-IDF value as '12 .6286 'by setting the TF value for document B to' 5 'and the IDF value to' 2.52573 '.

The keyword detection unit 312 may determine the important word as the higher the TF-IDF value multiplied by the TF value and the IDF value, and extract the corresponding word as a keyword of interest.

When the keyword of interest is extracted by the keyword detector 312, the document reliability calculator 320 determines the reliability of the document including the keyword of interest.

The document reliability calculation unit 320 may include a reliability extraction unit 321 and a social behavior reliability detection unit 322.

The reliability extraction unit 321 extracts the reliability of the document by weighting the latestness of the document, time weighting the latest text, and considering the number of positive responses ('Like'), the number of comments, and the number of shares left in the document do.

Equation 3 is an equation for calculating a weight for the reliability of a document left by a user on a social network.

로서, 사용자가 남긴 문서의 시간 가중치로 작성한 지 오래된 문서일수록 낮은 가중치를 가지게 된다. Like_w는 사용자가 남긴 문서의 긍정적인 반응(좋아요)의 수의 평균값이고, Comment_w는 사용자가 남긴 문서 평균값, Share_w는 사용자가 남긴 문서의 공유 평균값이다. 가중치 α, β, γ, δ의 합은 1로 표현된다.In Equation 3, Time_w is

As, the older the document is written as the time weight of the document left by the user, the lower the weight. Like_w is the average value of the number of positive responses (likes) of documents left by the user, Comment_w is the average value of the documents left by the user, and Share_w is the average value of the documents left by the user. The sum of the weights α, β, γ, and δ is expressed as 1.

The social behavior reliability detection unit 322 analyzes a user's social behavior and calculates user confidence. For example, the social behavior reliability detection unit 322 detects social behavior reliability based on at least one of the time when the user's social behavior was performed, the number of positive responses ('Like'), the number of comments, and the number of sharing times. You can.

The index calculator 330 combines social behavior reliability and TF-IDF techniques to finally calculate the interest index of the user terminal and determines the interest field.

Equation 4 is a formula describing a method for deriving a user's interest index.

In Equation 4, dn is the document left by the user on the social network, and dnT _i is the reliability of the document i left by the user on the social network. TF _j is a value indicating how often the word j appears in the document by word frequency. IDF is a value indicating the number of documents having a specific word in the entire document at the frequency of reverse documents.

The index calculator 330 sorts keywords of interest in ascending order according to keyword scores obtained in Equation 4 to extract top-n keywords.

However, if the user interest index is calculated using only the keyword scores made in this way, the amount of documents is different for each keyword, so in the case of the latest keyword, a user with a lower score than the old keyword may be recommended as an expert. In consideration of this problem, the index calculator 330 may calculate the interest index for each keyword by dividing the keyword score by the sum of user terminals having the same interest field on the social network.

4 is a block diagram showing a detailed configuration of a personal relationship index calculator in an expert recommendation system according to another embodiment of the present invention.

Referring to FIG. 4, the human relation index calculating unit 212 reconstructs the personal relations of the user terminals for each field of interest to calculate the human relation index.

The human relationship index calculating unit 212 may include a human network rebuilding unit 410 and an index calculating unit 420 according to interests.

The human network re-establishment unit for each interest field 410 analyzes a user's activity on the social network system and reconstructs the human network for each interest field with the result of extracting the interest field for each user terminal.

The index calculation unit 420 calculates a personal relationship index (PageRank score) for each field of interest by applying a selected algorithm (eg, 'PageRank') to the reconstructed personal relationship.

Here, the PageRank algorithm is an algorithm that calculates a score per web page through a link relationship between web pages, and the index calculation unit 420 applies the PageRank algorithm to a personal relationship to connect the user terminals of the same interest field. Can count.

For example, the index calculation unit 420 calculates a high PageRank score for an expert who has a personal relationship with many users in the same field, and a low PageRank for an expert who has low influence due to a small relationship with users in the same field. The score can be calculated.

An example in which the index calculation unit 420 rebuilds the human network for each field of interest will be described in detail in FIG. 6 to be described later.

5 is a block diagram showing a detailed configuration of a response quality index calculation unit in an expert recommendation system according to another embodiment of the present invention.

Referring to FIG. 5, the response quality index calculation unit 213 calculates a response quality index by analyzing past responses of each user terminal, so as to recommend an expert who can provide a quick and accurate answer to the querying user, Evaluate the response quality of experts in each field.

The response quality index calculation unit 213 may include a user response similarity analysis unit 510, a user response response analysis unit 520, and an index calculation unit 530.

The user response similarity analysis unit 510 measures the similarity between the interest field of the user terminal and the past response of the user terminal.

The user response similarity analysis unit 510 may calculate the similarity between the expert's interest field and the response text using cosine similarity.

Cosine similarity refers to the degree of similarity between vectors measured using the cosine value of the angle between two vectors in the inner space, and is frequently used for measuring similarity in a multidimensional positive space.

For example, the user response similarity analysis unit 510 organizes each keyword into each dimension and sets the document as a vector value expressed as the number of times each word appears in the document through cosine similarity in such a multidimensional space. Similarity between documents and keywords can be measured.

Equation 5 below calculates the cosine similarity, A is a query keyword, B is a document weight and is calculated as a vector value.

The user response response analysis unit 520 responds according to Equation 6 below based on the number of positive responses ('Like') and the time taken to respond to the past response of the user terminal on the social network. Determine the quality.

In Equation 6, Ar is the user's response amount, An is the user's answer n, Sim_An is the user's answer n's similarity of interest, likeAn is the user's answer n's average average value, and Atn is the user's average response time. The sum of the weights α and β is represented by 1.

The index calculating unit 530 calculates a response quality index by adding the similarity and the response quality.

6 is a diagram illustrating an example of rebuilding a human network between user terminals according to interests in an expert recommendation system according to embodiments of the present invention.

Referring to FIG. 6, the expert recommendation system may calculate a human relationship index after rebuilding a human network of a user terminal for each interest field.

6 (ⅰ) shows a human network that has been established between user terminals using a social network system. As shown in the figure, the existing human network is composed of user terminals having various interests A and B.

The expert recommendation system excludes user terminals that have registered interest field B from the existing human network shown in (i) of FIG. 6 to calculate the human relationship index of user terminals that have registered interest field A, and excludes interest area A After rebuilding the human network as shown in (ii) of FIG. 6 using the registered user terminals, the human relationship index may be calculated using the PageRank algorithm.

Likewise, the expert recommendation system excludes the user terminals registering interest area A from the existing human network shown in (6) of FIG. 6 in order to calculate the human relationship index of the user terminals registering interest area B, and the interest areas After rebuilding the human network as shown in (6) of FIG. 6 by using the user terminals that have registered B, the human relationship index may be calculated using the PageRank algorithm.

The expert recommendation system can determine the human network index for each interest by using the PageRank algorithm based on the human network reconstructed for each interest.

In addition, the expert recommendation system may calculate a personal relationship index for each field of interest based on Equation 7 below.

In Equation 7, PR (U ₂₎ is the keyword score value of the user U ₂ area of interest i, C (U ₂₎ is in the interests of the user U ₂ i friends, N is the total number of users with interest i Is a number, d is a dumping vector set between 0 and 1, and may be set to 0.85 in this specification.

As described above, the expert recommendation system can recommend a user with high influence and expertise as an expert among users with interests related to the query keyword, based on the personal relationship index for each field, so that the user has high influence and expertise.

7 is a diagram illustrating an example of providing a recommendation result to a group of experts matching a user query in the expert recommendation system according to embodiments of the present invention.

Referring to FIG. 7, the expert recommendation system according to embodiments of the present invention generates a tree 710 including a group of experts configured with respect to a query keyword and a group of sub-experts matching the lower keywords of the query keyword, , Each expert group can be arranged according to the hierarchical structure of the matched keyword. The expert recommendation system may output the expert recommendation result 720 based on the tree 710.

As an example, the expert recommendation system compares and matches the query keyword extracted from the user query with the tree 710 of FIG. 7 to search for expert group B matching the query keyword, and includes all user terminals belonging to the expert group B The expert recommendation result 720 may be output.

In addition, the expert recommendation system searches the sub-expert groups C, D, E branching from the expert group B in the tree 710 of FIG. 7, and further includes a user terminal belonging to each sub-expert group C, D, E. The recommended result 720 may be output.

Hereinafter, FIG. 8 describes in detail the workflow of the expert recommendation system according to embodiments of the present invention.

8 is a flowchart illustrating a procedure of a method for recommending experts according to an embodiment of the present invention.

The expert recommendation method according to this embodiment may be performed by the expert recommendation system described above.

Referring to FIG. 8, in step 810, the expert recommendation system checks whether a query request is received from a user terminal through a social network system.

When a query request is received in step 810, the expert recommendation system extracts a query keyword from the query in step 820.

For example, the expert recommendation system may extract the query keyword ('figure') by analyzing the query when a query is requested from the user terminal through the SNS ('When will the One Piece 6 rupee figure be released?').

In step 830, the expert recommendation system searches for an expert group matching the query keyword in the expert tree generated by considering the hierarchical structure of words.

For example, the expert recommendation system may search for a group of experts matching the query keyword in a tree maintained in the database.

In addition, the expert recommendation system may further search the sub-expert group diverging from the expert group in the tree.

In step 840, the expert recommendation system outputs expert recommendation results including experts belonging to the expert group.

In addition, the expert recommendation system may further output expert recommendation results by further including a sub-expert group diverging from the expert group.

For example, referring to FIG. 7, when the query keyword extracted from the user query matches the expert group B, the expert recommendation system includes the expert group B and the expert groups C and D, which match the subkeywords of the query keyword. Expert recommendation results can be output, including E.

At this time, the expert recommendation system includes some user terminals (higher experts) corresponding to the top ('1/3') in the order of expert index according to the number of each sub-keyword ('3') in each of expert groups C, D, and E. ) Can be selected and included in expert recommendation results.

In this way, by recommending an expert terminal capable of responding to user inquiries quickly and accurately, considering social behaviors such as posts, comments, sharing, and positive responses and interests of each user terminal and human network and response quality determined through human network analysis , It can increase the utilization of the Q & A service through SNS and user satisfaction.

The method according to an embodiment of the present invention may be implemented in the form of program instructions that can be executed through various computer means and recorded on a computer readable medium. The computer-readable medium may include program instructions, data files, data structures, or the like alone or in combination. The program instructions recorded on the medium may be specially designed and configured for the embodiments or may be known and usable by those skilled in computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical media such as CD-ROMs, DVDs, and magnetic media such as floptical disks. -Hardware devices specially configured to store and execute program instructions such as magneto-optical media, and ROM, RAM, flash memory, and the like. Examples of program instructions include high-level language code that can be executed by a computer using an interpreter, etc., as well as machine language codes produced by a compiler. The hardware device described above may be configured to operate as one or more software modules to perform the operations of the embodiments, and vice versa.

As described above, although the embodiments have been described by a limited embodiment and drawings, those skilled in the art can make various modifications and variations from the above description. For example, the described techniques are performed in a different order than the described method, and / or the components of the described system, structure, device, circuit, etc. are combined or combined in a different form from the described method, or other components Alternatively, even if replaced or substituted by equivalents, appropriate results can be achieved.

Therefore, other implementations, other embodiments, and equivalents to the claims are also within the scope of the following claims.

100: expert recommendation system
110: keyword extraction unit 120: terminal search unit
130: group configuration unit 140: result output unit
150: collection unit 160: interest register
170: network reconstruction unit 180: response quality evaluation unit
190: database

Claims (15)

Analyzing social behavior data collected from the social network system, and registering a field of interest for each of the user terminals;
Separately rebuilding the first human network on the social network system constructed between the user terminals into a second human network between user terminals having the same interest;
Extracting a query keyword from the query requested from the first user terminal;
Searching for a second user terminal in which the number of connected terminals is greater than or equal to a set value in a second human network reconstructed with respect to the field of interest associated with the query keyword;
Configuring the second user terminal as a group of experts capable of answering the query;
Matching the configured expert group to the query keyword;
Matching a sub-expert group consisting of each sub-keyword for the query keyword to the sub-keywords;
Generating a tree by arranging each of the expert group and the sub-expert group as a node in consideration of the hierarchical structure between matched keywords;
When the same query keyword as the above query is extracted from the new query,
Searching for a sub-expert group diverging from the expert group matching the query keyword in the tree; And
Outputting some user terminals selected in the sub-expert group as expert recommendation results to the user terminal requesting the new query according to the number of sub-expert groups diverging from the expert group.
Expert recommendation method for providing a question and answer service based on a social network system including a. delete According to claim 1,
Based on the time when the query is requested, social behavior data of each user terminal recorded including a text form document of at least one of posts and comments created or shared by each user terminal during a predetermined period of time, the social network Steps to collect from the system
Further comprising,
The step of registering the interest field,
Assigning a keyword score to each keyword of interest extracted from the document according to a term frequency-inverse document frequency (TF-IDF) technique;
Assigning a reliability score to each of the documents in which the keyword of interest is extracted, in consideration of the number of times the reaction is input and the date and time of writing the document;
Selecting a certain number of keywords of interest in the order of the keyword score and the reliability score; And
Registering the selected interest field keyword as an interest field of each user terminal
Expert recommendation method comprising a. According to claim 3,
The expert recommendation method,
Re-collecting, from the social network system, social behavior data of each user terminal recorded during the predetermined period of time based on the elapsed time, when a predetermined time has elapsed from the time when the query is requested; And
Updating the registered interest by analyzing the re-collected social behavior data
Expert recommendation method further comprising a. delete According to claim 1,
In the second human network re-established with respect to the field of interest associated with the query keyword, the first step of searching for the second user terminal connected to the first user terminal
Expert recommendation method further comprising a. According to claim 1,
In the case of a user terminal including a query response history in the social behavior data,
As a step of evaluating the response quality for the past response submitted by the user terminal, considering at least one of the similarity between the past response and the field of interest, the response time and the response of the other user terminal to the past response, Quantifying and evaluating the response quality; And
Searching for the second user terminal in the order of the numerical value indicating the response quality in the second human network reconstructed with respect to the field of interest associated with the query keyword
Expert recommendation method further comprising a. According to claim 1,
Identifying a subject according to the context of the query, and searching for the second user terminal that has registered an interest field associated with the query keyword for each subject
Expert recommendation method further comprising a. delete An interest field registration unit that analyzes social behavior data collected from the social network system and registers an interest field for each of the user terminals;
A network re-establishing unit that separately rebuilds the first human network on the social network system constructed between the user terminals into a second human network between user terminals having the same interest;
A keyword extraction unit that extracts a query keyword from the query requested from the first user terminal;
A terminal search unit searching for a second user terminal in which the number of connected terminals is greater than or equal to a set value in a second human network reconstructed with respect to the field of interest associated with the query keyword;
The second user terminal is configured as a group of experts capable of answering the query, the group of experts configured to match the query keyword, and a group of sub-experts each composed of sub-keywords for the query keyword, each sub A group constructing unit that matches a keyword and considers the hierarchical structure between the matched keywords to generate a tree by arranging each of the expert group and the sub-expert group as a node; And
When the same query keyword as the above query is extracted from the new query,
In the tree, search for a sub-expert group diverging from the expert group matching the query keyword, and according to the number of sub-expert groups diverging from the expert group, some user terminals selected in the sub-expert group , A result output unit that outputs as the expert recommendation result to the user terminal requesting the new query
Expert recommendation system for providing a question and answer service based on a social network system including a. delete The method of claim 10,
Based on the time when the query is requested, social behavior data of each user terminal recorded including a text form document of at least one of posts and comments created or shared by each user terminal during a predetermined period of time, the social network Collection part collected from the system
Further comprising,
The interest registration section,
For each keyword of interest extracted from the document, a keyword score is assigned according to the TF-IDF technique,
For each of the documents in which the keyword of interest is extracted, a reliability score is given in consideration of the number of times the reaction was input and the date and time of writing the document,
A certain number of interest keywords are selected in the order of the keyword score and the reliability score,
Registering the selected interest keyword as the interest of each user terminal
Expert referral system. delete The method of claim 10,
In the case of a user terminal including a query response history in the social behavior data,
When evaluating the response quality for the past response submitted by the user terminal, the response quality is considered by considering at least one of the similarity between the past response and the field of interest, the response time, and the response of the other user terminal to the past response Quality evaluation department that evaluates by digitizing
Further comprising,
The terminal search unit,
In the second human network re-established with respect to the field of interest associated with the query keyword, the second user terminal is searched in the order of the number representing the response quality.
Expert referral system. delete

Images