KR101811638B1

KR101811638B1 - Method of Influence Measurement based on Sentiment Analysis of SNS Users

Info

Publication number: KR101811638B1
Application number: KR1020160012886A
Authority: KR
Inventors: 양형정; 정회윤
Original assignee: 전남대학교 산학협력단
Priority date: 2016-02-02
Filing date: 2016-02-02
Publication date: 2017-12-27
Also published as: KR20170092168A

Abstract

The present invention relates to a method for measuring the influence of a user of a social network service on another user of a social network service, and a method of measuring an influence by an emotional analysis of a user of the social network service according to an embodiment of the present invention, Collecting and pre-processing the data disclosed in Fig. Processing the preprocessed data in natural language; Extracting data included in a predetermined emotion list from the natural language processed data; Calculating a user's emotional index by applying the extracted data to a predetermined polar label, and analyzing a user's emotional state using the emotional index; And measuring the influence of the data creator disclosed on the social network service using the analyzed user's emotions.

Description

METHOD FOR MEASURING INFLUENCE BY SENSITIVITY ANALYSIS OF SOCIAL NETWORK SERVICE USERS [0002]

The present invention relates to a method of measuring the influence of a social network service user on other social network service users, and more particularly, to a method of analyzing a social network service user by merging emotional elements of a social network service user into a structural element of the social network service, The present invention relates to a method for measuring an influence by a user's emotional analysis of a user of a social network service that measures a user's influence on other users of a social network service.

The modern society is a social network community where information exchange between the members of society is constantly carried out due to the spread of the Internet and smart phones. The development of such a society has played a major role in the use of diverse social media such as Twitter, Facebook, Instagram as well as Internet devices such as smart phones. As the use of social media grows, information that is constantly shared is constantly accumulating. As such information is used for public opinion formation and agenda setting, research on user influence on information propagation of social media is increasing.

Impact measurement and analysis play an important role in a variety of areas such as marketing, politics and advertising. In marketing, influence plays a role in promoting products and driving reputation. For politicians, it is an important predictor of publicity and electoral success and failure. In addition, impact measurement in the advertising field is important in terms of rapid information transmission and low cost, high efficiency information delivery.

Many existing impact measurement methods are viewed from the structural point of view of the SNS. In the case of influence measurement using Twitter, the influence of users was predicted by using only twitter structural information such as Follower, Retweet, Reply, Mention among users.

However, empirical factors affecting decision making, such as information transmission, public relations, public opinion formation, and agenda setting, can be seen through studies that have different rates of diffusion for emotional words. This implies that emotional factors and information diffusion influence are closely related.

Typical research to measure influence is to measure influence using centrality. There are various methods to measure the centrality, such as the center of the interaction, the center of mediation, and the centrality of the connection. Closeness Centrality measures the shortest path between two different users and classifies the user with the smallest sum of the shortest paths as the most influential user in the whole network.

"Twitter Data Analytics" (Shamanth Kumar, Fred Morstattter, Huan Liu, "Twitter Data Analytics", Springer, August 19, 2013), authored by Shamanth Kumar et al., Applied a "Betweenness Centrality" Is a measure of the degree to which a mediator plays a central role.

That is, the more the user is located on the shortest path, the higher the influence of the user. Degree Centrality is the sum of the points connected to a point in the network, with an emphasis on the extent to which they are connected to other points. The influence is measured as the ratio of the degree of inward connection and the degree of outward connection of each actor in the total number of connections. The influence measurement using the structural centralizability has a problem that it can not exceed the limit considering only the structure.

In other words, there is a disadvantage that it is difficult to express the problem of the real world which expresses various social relations in that the correlation of users is measured by merely structural characteristics.

Another typical impact measurement algorithm is the PageRank algorithm applied to Google search. (Park, Ji-Hye, ServoMill, The Influence of Powerful Mediators on Online Social Network Service Environment, Journal of information technology applications & management)

Page rank is weighted according to relative importance to measure influence. PageRank is applied to any group that is cited with reference to each other. However, due to the nature of social network services that exist in the real world, a small number of opinions can provide a higher influence than many opinions.

Short and Tweet: Experiments on Recommendation (Jilin Chen, Rowan Nairn, Les Nelson, Michael Bernstein, Ed H. Chi, and Rowan Nairn, "Short and Tweet: (CHI '10), pp.1185-1194, 2010). In order to address the hypothesis that many followers are influential on the Internet, It is a research that started. In this study, we collected the entire data of Twitter in 2009 to verify the scientific data base.

The number of followers, the number of responses to tweets, and the number of retweets were measured. As a result of measuring the influence by using three scales, it was confirmed that the rank of influence is not constant but varies according to the scale.

As a result of the data analysis, the relationship between followers and mentions or retweets showed a low correlation. In other words, the follower is a popular Twitter user, not necessarily mentioning or retransmitting. In other words, popularity and influence are different.

Therefore, in order to solve these problems, it is necessary to study influence measurement considering emotional similarity among users.

SUMMARY OF THE INVENTION The present invention has been made in view of the above problems, and it is an object of the present invention to provide a social network service which can more accurately measure influence on an influencer or an opinion leader, And to provide a measurement method.

In order to solve the above object,

A method for measuring an influence by an emotional analysis of a user of a social network service according to the present invention includes collecting and pre-processing data disclosed on a social network service; Processing the preprocessed data in natural language; Extracting data included in a predetermined emotion list from the natural language processed data; Calculating a user's emotional index by applying the extracted data to a predetermined polar label, and analyzing a user's emotional state using the emotional index; And measuring the influence of the data creator disclosed on the social network service using the analyzed user's emotions.

According to an aspect of the present invention, the preprocessing step collects and stores the structural information of the social network service using the REST API.

According to one aspect of the present invention, the natural language processing step performs morpheme analysis including root analysis, root extraction, and word segmentation after eliminating an idiomatic word from the preprocessed data.

According to one aspect of the present invention, the emotion index is expressed in a tree structure using the Recursive Neural Tensor Network and the extracted data is expressed in a tree structure using a predefined emotion dictionary The emotion index is calculated by mapping the emotion value to the data.

According to one aspect of the present invention, the step of analyzing the influence can be quantified and analyzed by the following Equation (18).

&Quot; (18) "

here,

Is an influence value considering user's emotions,

Is the weight set by the measurer, R is the probability of being transmitted to tweet j through tweet over tweet i,

Is a weight considering emotion.

On the other hand,

Is defined by Equation (2) below.

&Quot; (17) "

here,

The

ego,

Is a weight set by the measurer, k is an arbitrary value greater than 0,

Is the emotion similarity between the user and the data creator, U is a set of users, and u is any user.

On the other hand,

Is defined by Equation (14) below.

&Quot; (14) "

here,

Wow

Is a hash tag emotion analysis result set that user i and user j possess.

According to the method for measuring the influence by the emotional analysis of the user of the social network service according to the present invention,

In addition to the structural elements of social network services, it is possible to measure the influence in consideration of the emotional similarities among the users of social network services, so that the influencers or Opinion Leaders that have a great influence in social network services It is possible to more precisely measure the influence.

In addition, by accurately measuring the influence of Influential or Opinion Leader, it can be used to transfer information quickly and efficiently at low cost in various fields such as politics, marketing, and advertisement.

1 is a diagram showing a configuration of a social network service providing system.
FIG. 2 is a flowchart illustrating a method of measuring influence by emotional analysis of a user of a social network service according to an exemplary embodiment of the present invention.
3 is a flowchart illustrating a preprocessing step of a method of measuring influence by emotional analysis of a user of a social network service according to an embodiment of the present invention.
4 is a flowchart illustrating a natural language processing step of an influence measurement method by emotion analysis of a user of a social network service according to an embodiment of the present invention.
5 is a flowchart illustrating an emotional analysis step of a method of measuring influence by emotional analysis of a user of a social network service according to an exemplary embodiment of the present invention.

The present invention discloses a method for measuring the influence of information delivery on a social network service (e.g., a twitter). For influence measurement, the present invention collects Twitter data, performs natural language processing and emotion analysis. The analyzed information is used to calculate the probability of being transferred from the user's tweet to the other tweet. Based on the probability information of the tweet, the relative importance is analyzed considering the retweet, follower, reply, Apply similarity weights to determine impact. In the present invention, a twitter, which is a kind of social network service, is exemplified and is not necessarily limited to twitter.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings. In the drawings, the same reference numerals as used in the appended drawings denote like elements, unless indicated otherwise. In the following description of the present invention, a detailed description of known functions and configurations incorporated herein will be omitted when it may make the subject matter of the present invention rather obvious or understandable to those skilled in the art.

1 is a diagram showing a configuration of a social network service providing system.

1, the social network service providing system 100 includes a plurality of terminals 10, a service apparatus 20, and a communication network 30.

A processor mounted on the terminal 10 or the service device 20 may process program instructions for executing the method according to the present invention. In one implementation, the processor may be a single-threaded processor, and in other embodiments, the processor may be a multi-threaded processor. Further, the processor is capable of processing instructions stored on a memory or storage device.

The communication network 30 performs a series of data transmission and reception operations for data transmission and information exchange between the plurality of terminals 10 and the service apparatus 20. [ In particular, the communication network 30 may use various types of communication networks and may be, for example, a wireless LAN (WLAN), a Wi-Fi, a Wibro, a WiMAX, HSDPA, and High Speed Downlink Packet Access) or a wireless communication method such as Ethernet, xDSL (ADSL, VDSL), Hybrid Fiber Coax (HFC), Fiber to the Curb (FTTC) A wired communication method can be used. The communication network 30 is not limited to the above-described communication system, but may include any other widely known or later-developed communication systems as well as the communication systems described above.

The terminal 10 communicates with the service device 20 through the communication network 30 and transmits and receives data for a social network service with another terminal. In particular, the terminal 10 according to the present invention executes a social network service with at least one other terminal.

The service apparatus 20 communicates with a plurality of terminals 10 through a communication network 30 to transmit and receive data for a social network service.

FIG. 2 is a flowchart illustrating a method of measuring influence by emotional analysis of a user of a social network service according to an exemplary embodiment of the present invention.

The method for measuring the influence by the emotional analysis of the user of the present invention includes collecting and analyzing the social network service data performed by the user through the social network service providing system of FIG. Measure influence. The method described below may be performed by a computing device that includes a processor capable of processing the collected social network service data.

As shown in FIG. 2, a method for measuring influence by an emotional analysis of a user of a social network service according to an embodiment of the present invention includes data collection and preprocessing step (S100), processing of the preprocessed data in a natural language (S300) of extracting data included in a predetermined emotion list from the natural language processed data (S300), analyzing emotion of a user (S400), and influencing analysis step (S500). Each of these steps may be performed by a computer device or the like programmed to do so.

The data collection and preprocessing step S100 collects and preprocesses data disclosed on the social network service. For example, the data disclosed on the tweeter is collected and preprocessed.

For this purpose, as shown in FIG. 3, first, twitter data including a hash tag is collected (S110)

Here, the data on the tweeter is collected using the REST (Representational State Transfer) API (see https://dev.twitter.com/rest/public).

Next, user information corresponding to the tweet is collected from the Twitter data collected using the REST API (S120)

Next, the number of followers of the tweet is counted and it is judged whether or not it exceeds a predetermined reference value. (S130). In the figure, it is determined whether the number of followers exceeds 20 by a predetermined reference value. If the number of followers exceeds 20, it is regarded as meaningless data and deleted (S140). If the number exceeds 20, the user information is stored (S150)

Next, the user's timeline is collected using the REST API (S160), the timeline information is retrieved (S170), retweet, mentions, etc. are collected (S180) and stored (S190)

Here, the time line is an interface for collecting and displaying the user's own and follower's posts on a social network service site (e.g., Twitter). In general, the user's own text is displayed, and the follower's text is displayed And writes to users.

On the other hand, since REST API has access to past tweets, there are various kinds of search conditions that can be defined. The information collected using the REST API consists of a structure that collects desired information by reading a web page made of XML (Extensibility Generating Language) file. The user's timeline, follower, list, user information, account information , Location information, and the like.

Next, the natural language processing step (S200) will be described with reference to FIG.

4 is a flowchart illustrating a natural language processing step of an influence measurement method by emotion analysis of a user of a social network service according to an embodiment of the present invention.

As shown in FIG. 4, the natural language processing step (S200) performs natural language processing using the collected tweeter data in the preprocessing step (S100). Since the collected Twitter data are natural language languages used in everyday life, NLP (Natural Language Processing) process is required for computer processing.

For natural language preprocessing of Twitter data, general terms such as stemming, root extraction, and word separation are extracted for each user's tweet information. In this process, the frequency of occurrence of each term is additionally stored for the emotional analysis based on the hash tag.

First, the twitter data collected in the preprocessing step is input for the natural language processing (S210). The inputted twitter data can be collected from the user's timeline using the above-described REST API, and stored in the retrieve, mentions, and the like.

Next, an idiomatic word included in the collected tweeter data is removed (S220). An idiomatic word is meaningless words such as articles, prepositions, surveys, and conjunctions.

Removal of such an idiomatic word can be eliminated by using an idiomatic language table shown in Table 1 below.

Index StopWord One a 2 about 3 above 4 across 5 after 6 again 7 against 8 all 9 almost 10 alone 11 along 12 already

Next, morphological analysis is performed on the twitter data from which the stopwords are removed (S230). Morphological analysis is a general term extraction process such as root analysis, root extraction, and word separation.

For example, in the case of nouns, root analysis works in the form of singular, plural, but it means the process of converting the same into different forms depending on the number. In the above case, the conversion process is converted to the basic type.

Here, root extraction is also called stemming, and it is a task to remove special symbols or stopwords of research. The word separation process is to remove words from the changed word and find the original shape of the word.

Through the above process, a natural language processing data table composed of terms and appearance frequencies is created from the tweet set.

Next, data included in the default emotion list is extracted from the natural language processing data created through the above process (S300). That is, the tweeter data is separated into morpheme units and processed in a natural language, and among the natural language processed tweeter data, Extracts the twitter data included in the emotion list of the user, and labels the extracted twitter data. An example of such an emotion list and labeling is shown in Table 2 below.

Also, the degree of emotion is analyzed by using Twitter message and emotion list classified as morpheme. This analyzes the polarity of the overall Twitter message by determining whether or not there is a word contained in the sentence vocabulary list. As shown in FIG. 6, the closer to blue, the more positive, and the closer to red, the more negative.

Next, the user's emotional analysis step (S400) will be described with reference to FIG. 5 is a flowchart illustrating an emotional analysis step of a method of measuring influence by emotional analysis of a user of a social network service according to an exemplary embodiment of the present invention.

5, first, a polarity label is configured for the user's emotional analysis (S410). The polarity label includes all polar adjectives having polarity extracted from the tweeter data collected in the natural language process S200 , And labeling the state name.

The emotion index of the user is calculated by applying the data extracted in step S300 to the polarity label (S420)

For example, Stanford Sentiment Treebank (Richard Socher, Alex Perelygin, Jean Wu, Jason Chuang, Christopher D. Manning, Andrew Ng, and Christopher Potts. 2013. Recursive deep models for semantic compositionality over a sentiment treebank.In EMNLP 2013, pages 16311642.) to analyze emotions. This method expresses each word in a tree structure using Recursive Neural Tensor Network and maps each emotion value using a predefined emotion dictionary. That is, for example, the emotion information can be classified using the API having the emotion category provided by Stanford as described above, and the degree of emotion expressed in each category can be analyzed.

Next, the influence of the data creator disclosed on the social network service is measured using the analyzed user's emotions (S500). For example, if a specific Twitter creator is influential or influential on Twitter, (Opinion Leader). To this end, the present invention proposes a method capable of quantitative determination through tweeter modeling.

Twitter modeling

On Twitter, there are various ways of exchanging information among Twitter users. Users can check their follower's follow-up relationship with other users, or follow-up users can reweet other users' tweets or create replies to create a third person Can receive the tweet.

In the present invention, the tweeter is modeled as follows to show the influence of the user on the tweeter. In the collected data, the set of tweets is defined as T, the set of users as U, and the set of hash tags as HS.

Retweet (RT) is a case in which a tweet j is created on the timeline of another user who has tweeted up the tweet i and expressed as Equation (1).

The reply (Reply, RP) is as shown in Equation (2) when a user creates a reply from the Tweet i to generate a Tweet j.

The RT (Equation 1) can be expressed as | T | x | T | (Math 2) will only display the value in the matrix of | T | x | T | Size matrix. In addition, the difference in value is due to differences in retweets and comments.

Mention (MN) is a case in which a tweet i is created as a mentee to a user j, as shown in Equation (3).

Following (FW) is a case in which user i follows user j, which is expressed as Equation (4).

The hash tag (HashTag, HT) is created using the hash tag (HS) in the tweet i, and is expressed by Equation (5).

Twitter impact measurement

On Twitter, the tweets you create are sent to the users around you via retweet, mentions, and replies. However, it is difficult to accurately calculate the influence if you only take actions such as retweet, mentions, and replies, which are actions that are transmitted directly through the link, due to the structure of the sparse connectivity between tweets on Twitter.

Therefore, the present invention considers not only the relationship of tweets, but also the way in which tweets are transmitted, and the influence is measured.

The influence calculation method considering only the structural factors without considering the emotional factor is shown in Equation (6). We define the probability that information will be delivered through the user's behavior through Equation (6).

In Equation (6)

Can be divided into the probability of accessing the tweet at random and the case of accessing through activity on Twitter. Also,

The value of

Weights are given to each. These values can be arbitrarily set by the measurer, but it is preferable that the sum thereof be 1. This is to prevent the weight value of each element from becoming too large or small. The probability calculations from each twitter activity are as follows.

The tweets you create are sent to other tweets via retweets or responses. Using Equation (1) and Equation (2), the probability of being transmitted from tweets i to tweets j

) Is defined as Equation (7).

M for all mentions about tweets created by user i, and mn for mentions of user j. At this time, regarding the probability that user j will make a mention for the tweet created by user i

. Thus, the probability that the tweet will be propagated to the user j by mentioning the user about the tweet created by the user i

.

percentage

Can be expressed as Equation (9) using Equation (8).

Is the number of tweets that have mentions in tweets.

The

Of the total number of tweets.

The probability that a user accesses user j through i's followers

This approach allows you to access your j's tweets. The probability that a tweet i accesses a user j through the follower information of the user i and accesses the tweet j

Can be expressed by Equation (11) using Equation (10).

Is the number of followers of user i.

The

Of the total number of tweets.

Users with a common theme across tweets often access more often than users with different themes. An expression for defining whether users i and j have similar themes (

Can be expressed as Equation (13) using Equation (12).

Is the number of tweets containing the hash tag k,

Is the total number of hash tags included in tweet i.

The probability of being delivered when the emotional similarities of user i and user j are similar (

) Uses the common hash tags of users to define what emotions users have about a subject.

Assuming that a set of emotional analysis results for a hash tag common to users i and j is hs, the similarity degree of emotion between users is defined as Equation (14) through the cosine similarity.

Here, hsi is a set of hash tag emotion analysis results possessed by user i, and hsj is a set of hash tag emotion analysis results possessed by user j, the similarity degree of emotion between users is expressed by Equation (14) define.

Equation (14) is used to apply the weight for emotion to the influence measurement method defined in Equation (6) as Equation (15).

Equation (15) is a method of measuring influence among users through retweeting, mentions, following, and hash tags. Applying emotional weighting to Equation (15) through Equation (6), an emotional element influence measurement method is applied.

&Quot; (14) "

&Quot; (6) "

&Quot; (15) "

The influence G of Equation (6) can be expressed by Equation (16) using Z in Equation (15).

In Equation (16), Retwitt, Mention, and Reply, which affect emotional similarity between user i and j, apply a weight to emotionally similarity in Z. [ Because emotional similarity affects the probability of delivery to a user's behavior, we only apply weights to Z. The method of applying the emotional weight is defined as Equation (17). We analyze each other's weight information about each subject about user i, j.

Equation (17) represents the emotional weight of the user j for the emotional weight for all users. The emotional weight value of user i for user j is weighted on the emotional factor of existing retweet, mentions, followers, and hash tags. The constant k is greater than zero to eliminate computational errors for the degree of similarity of zero.

The constant k is an arbitrary value greater than 0 in order to prevent false results from occurring when the similarity result is zero. Finally, the influence G considering emotion is expressed by Equation (18).

Equation (18) can be used to accurately measure influence on an Influential or Opinion Leader on a social network service by using an influence measuring method that takes an emotional factor into account in the structural elements of Twitter have.

According to the method for measuring influence by the emotional analysis of the user of the social network service according to the present invention, the influence can be measured not only in terms of the structural elements of the social network service, but also in consideration of the emotional similarity among the users of the social network service, It is possible to more accurately measure the influence of Influential or Opinion Leader on social network services. In addition, by accurately measuring the influence of Influential or Opinion Leader, it can be used to transfer information quickly and efficiently at low cost in various fields such as politics, marketing, and advertisement.

While the present invention has been particularly shown and described with reference to exemplary embodiments thereof, it is clearly understood that the same is by way of illustration and example only and is not to be taken by way of limitation in the spirit and scope of the invention as defined by the appended claims. It will be understood by those skilled in the art that various changes in form and detail may be made therein without departing from the scope of the appended claims.

10: Terminal 20: Service device
30: communication network 100: social network service providing system

Claims

A computing device, comprising: a computing device,
A preprocessing step of collecting SNS data shared among users based on a structural element of a social network service (SNS);
A natural language processing step of processing the SNS data by natural language processing to extract words included in the SNS data and appearance frequencies of the words;
Extracting a word included in a predetermined emotion list among words extracted by the natural language processing;
Calculating a user's emotion index for each SNS data by applying the extracted word to a predetermined polarity label and analyzing emotions of the user using the emotion index; And
Calculating the emotion similarity between the user who created each SNS data and the other user who has received the SNS data by using the emotion of the analyzed user, calculating the SNS data on the SNS in consideration of the structural elements of the SNS and the calculated emotion similarity, Analyze the impact of the user you created
A method of measuring influence by emotional analysis of a user of a social network service including

The method according to claim 1,
Wherein the pre-processing step includes collecting and storing the SNS data using the REST API.

The method according to claim 1,
Wherein the natural language processing step performs morpheme analysis including at least one of root analysis, root extraction, and word segmentation after eliminating the idioms from the SNS data. Way.

The method according to claim 1,
The emotion index expresses the extracted words in a tree structure using a Recursive Neural Tensor Network and maps emotion values to words expressed in a tree structure using a predefined emotion dictionary And calculating the influence by the emotional analysis of the user of the social network service.

The method according to claim 1,
Wherein analyzing the influence comprises:
And calculating influences in consideration of the structural elements of the SNS and emotion similarity among users according to Equation (18) below.
&Quot; (18) "