Research Article Microblog Sentiment Orientation …downloads.hindawi.com/journals/jece/2016/7282913.pdfclassify its sentiment polarity. 3. Problem Definition. . Motivation. e purpose

Research ArticleMicroblog Sentiment Orientation Detection Using UserInteractive Relationship

Liang Wang Mei Wang Xinying Guo and Xuebin Qin

Xirsquoan University of Science and Technology Xirsquoan 710054 China

Correspondence should be addressed to Liang Wang wangliangxusteducn

Received 26 November 2015 Accepted 16 February 2016

Academic Editor Jiang Zhu

Copyright copy 2016 Liang Wang et al This is an open access article distributed under the Creative Commons Attribution Licensewhich permits unrestricted use distribution and reproduction in any medium provided the original work is properly cited

The development and popularity of microblog have made sentiment analysis of tweets and Weibo an important research fieldHowever the characteristics of microblog message pose challenge for the sentiment analysis and mining The existing approachesmostly focus on the message content and context information In this paper we propose a novel microblog sentiment analysisframework by incorporating the social interactive relationship factor in the content-based approach By exploring the interactiverelationship on social network based on posted messages we build social interactive model to represent the opposition oracceptation behavior Based on the interactive relationship model the sentiment of microblog message with sparse emotion termscan be deduced and identified and the sentiment uncertainty can be alleviated to some extent Afterwards we transform theclassification problem into an optimization problem Experimental results on Weibo data set indicate that the proposed methodcan outperform the baseline methods

1 Introduction

With the advent of Web 20 users become more eager topublish and share their opinions in various domains onsocial networks such as Twitter and Weibo The opinionscan reflect peoplersquos sentiments and views across areas asdiverse as commercial products services and public eventswhile these opinions can influence the ultimate decisionalprocess related to individual behavior and public policy Forexample by analyzing the sentiment of consumers for acertain brand and released product it can help companiesto improve their marketing campaign product design anduser experience After comprehending mass votersrsquo opinionstowards a candidate it is conducive to making their politicalstrategy better [1ndash3] So as it were the microblog sentimentanalysis has become a hot area which attracted attentions inboth academic and industrial fields

However the microblog message is usually short noisyand ambiguous due to its informal expression style Thesecharacteristics present a challenge for the text sentiment anal-ysis Recently many works are making great efforts to dealwith the microblog sentiment classification problem Gener-ally speaking the previous works are mostly based on the

content-based approach such as lexicon-based method andcorpus-based method [4ndash6] The content-based approachneeds to establish a sentiment word dictionary beforehandAnd through the sentiment word matching calculation thecorresponding sentiment could be labeled to the microblogmessage to be analyzed Actually the previous researchesmostly rely on predefined sentiment vocabularies which arehighly domain specific

The existing research works have proved that the content-based method is effective in tackling microblog messageswith strong sentiment expressions However sometimes themessage does not contain obvious sentiment terms butthere still exists emotional tendency implicitly to a certainextent For this situation the pure content-based approachis inadequate to identify the sentiment orientation In otherwords the sparse sentiment terms in microblog messagesmake it difficult to classify the hidden emotion Moreoverin the microblog message dissemination process the sen-timent characteristic may not be discriminative to classifyits emotion polarity In view of this situation some scholarsconsider incorporating the hidden social relationship suchas friend and follower into message content to settle the

Hindawi Publishing CorporationJournal of Electrical and Computer EngineeringVolume 2016 Article ID 7282913 6 pageshttpdxdoiorg10115520167282913

2 Journal of Electrical and Computer Engineering

sparse sentiment terms problem [7 8] The research worksin [7 8] are based on the hypothesis that people with friendrelationship often keep similar sentiment towards a giventopic

Meanwhile in practice people may be in disagreementwith others in their friend circle as they have differentsocial cognition knowledge structure and experience Thatis to say one may oppose or accept the opinion reflectedin message posted by someone else And the oppositionor acceptation behavior also reflects the emotion polarityof the participants in the microblog environment In thispaper this opposition or acceptation behavior is regardedas a social interactive relationship By leveraging this kindof interactive relationship it may be feasible to infer thesentiment of message for participants and build the learningmodel In this work we exploit the social relationshipbetween users on microblog platform and incorporate thisfactor into content-based analysis process Via exploitingthe social interactive relationship among microblog users avoting-based social interactive model has been built to linkthe microblog participants based on the message contextAfterwards the sentiment classification problem can betransformed into an optimization problem integrating thecontent-based strategy and social interactive relationshipfactor Experimental results on microblog message data setsvalidate the effectiveness and efficiency of our method

This paper is structured as follows In Section 2 wereview related work on sentiment classification After that wedescribe and formulate the social interactive relationship inmicroblog message and propose the classification model indetail in Section 3 In Section 4 we then evaluate our modelsand present experimental results Finally in Section 5 wesummarize our findings and contributions

2 Background and Related Works

In recent years as an opinion-rich resource microblogmessages provide the possibility of mining and analyzing theemotion of people in different scales And microblog senti-ment orientation identification has gained hung popularityand attracted researchersrsquo attention across many fields In thissection an overview of works related to microblog sentimentanalysis and classification is presented

Previous works usually compare the text content of agivenmessagewith a lexicon or a dictionary to classify its sen-timent and calculate its strength For instance SentiWordNetcontains as many as 200000 entries to match each word withpositive negative or objective scores Mostafa et al proposea lexicon-based method to analyzing consumersrsquo sentimenttowards some commercial brands [1 2] Go et al [2] studythe problem of sentiment classification for Twitter messagesby the means of machine learning algorithms Based ondistant supervised learning thework compares the sentimentclassification performance for three algorithms includingNaıve Bayes Maximum Entropy and SVM Ghiassi et al [3]introduce an approach to supervised feature reduction using119899-grams and statistical analysis to develop a Twitter-specificlexicon for sentiment analysis Kumar and Sebastian [4]

expound a hybrid approach to analyze the sentiment of Twit-terThe approach employs a corpus-basedmethod to identifythe semantic orientation of adjectives and a dictionary-basedmethod to find the semantic orientation of verbs and adverbsSaif et al [5] add semantics information as additional featuresin the process of tweets sentiment analysis Moreover thecorrelation of the representative concept is calculated withpositive and negative sentiment Bollen et al [6] extract sixdimensions of emotion using an established psychometricinstrument POMS and preform tweet sentiment analysis inthe field of stock market crude oil price indices and majorpublic events

As tweets are usually short and more ambiguous Jianget al [9] study the target-dependent Twitter sentiment clas-sification problem Through incorporating target-dependentfeatures and context information related to the given targettweet the sentiment of query tweet can be classified aspositive negative or neutral Pang and Lee [10] review therelated approaches for opinion mining and sentiment analy-sis Besides some other issues including privacy protectionmanipulation and economic impart are also discussed in[10] Althoughmany researchers point out that the propertiesof short and noisy present new challenges to the sentimentanalysis for microblog message Bermingham and Smeaton[11] find classifying sentiment in microblogs easier than inblogs Davis and OrsquoFlaherty [12] evaluate the automatedsentiment coding accuracy and misclassification errors of sixleading third-party companies for a broad range of commenttypes and forms

Actually beside the content-based information the hid-den social relationship in microblog message can also bereflected and may be used in the sentiment analysis processRecently there are many works that focus on the socialrelationship in microblog space and leveraged the relation-ship to improve the classification accuracy [7 8] Hu etal [7] exploit social relationship and introduce two socialtheory hypotheses into social network which are sentimentconsistency and emotional contagion Pang and Lee [10]expand the hypothesis verified by Hu et al in [7] andproposed a microblog sentiment classification frameworkInspired by the work in [7] we consider combining theimplicit social relationship behind the microblog message toclassify its sentiment polarity

3 Problem Definition

31 Motivation The purpose of sentiment orientation iden-tification for microblogs is to build a program which canautomatically identify whether a given microblog messageis expressing positive negative or no sentiment In otherwords this problem can be defined as follows given acollection of microblog message set 119863 and a set of classes119875119866 = positive neutralnegative the goal is to constructa mapping function Φ119866 119863 rarr 119875119866 which describes thesentiment orientation of messages according to an integratedmodel

However the character of microblog message such asnoisy short and ambiguous poses a challenge for sentimentanalysis and mining In addition the informal language

Journal of Electrical and Computer Engineering 3

expression and the use of emoticon make the sentimentanalysis problemmore difficult If the message to be analyzedcontains distinct sentiment vocabulary it is not hard toclassify the sentiment of this message by the means ofestablished semantic lexicon and dictionary However whenthere is no obvious sentiment vocabulary in the message butwith sentiment orientation it is impossible to obtain correctsentiment estimation sometimes it is difficult to determinethe implicit emotion via the content-based approach Espe-cially in the process of information dissemination such asforward comment and operations the original microblogmessage may change its sentiment to some extent and evengenerate polarity reversal Let us consider two examples asfollows

Example 1 Well Janersquos opinion towards genetically modifiedfood is so What do you think

Example 2 Jack I do not agree with your views aboutgenetically modified food

As described above in Example 1 the poster of thismessage does not express his opinion for the topic discussedby Jane But it is undeniable that the messagersquos holderalso delivers subtle emotion in the published message ForExample 2 the user claims that he disapproves the opinionof Jackrsquos viewpoint but the explicit sentiment towards thegiven topic remains unknown for analysts That is to say theuserrsquos emotion is opposite to the sentiment of Jackrsquos For thesetwo cases it is quite clear that context-based approach cannotobtain sentiment analysis results

32 Social Interactive RelationshipModel Obviouslymicrob-log message possesses not only explicit text content informa-tion but also implicit social interactive relationship due toits social network nature In microblog website a messageis published by a user to express hisher opinion towardsevent public figure and commercial products while thismessage may be forwarded and commented on by otherpeople who are interested in it Via the social interactiveoperation the opinion implied in a message diffuses invirtual cyber space and influences many participants It isquite clear that the social interactive operation plays a majorrole in the information dissemination and can reflect theemotion among the participants Moreover in the process ofmicroblog message diffusion the social interactive operationmay enhance or weaken or even reverse the emotion hiddenin the original message In other words social interactionis a key issue for dissemination and evolution of microblogmessage

In practice some people usually hold an objective andcomprehensive opinion in their published messages whilesome people may keep a subjective or extreme opiniontowards the evaluated objects such as events or other peopleAs a matter of common sense the posted messages ofthe former microblog users should obtain more supportor acceptance from other people And the latter ones willreceive criticisms for their one-sided opinions The supportor opposition can be seen as a vote from other people to

Carl

Jack

Tom

Jane

AgreeOppose

Figure 1 A toy example for social interactive relationship

the publisher of certain message If the pattern of socialinteractive relationship can be obtained it is possible to inferand predict the sentiment of microblog messages Based onthe observation we consider leveraging this kind of socialinteractive relationship in the process ofmicroblog sentimentclassification

Let us demonstrate the principle by a toy example inFigure 1 In the toy example there are four users TomJack Jane and Carl The social interactive relationship isrepresented by the directed line where the red one denotesagreement relation and the blue one is opposition relationAnd the direction of line is a forward comment or operation to diffuse microblog messages As in Figure 1 Tomhas opposed message published by Jane once and acceptedthe view of message posted by Carl onceThus assuming thatCarl has posted amicroblogmessage related to a certain topicif there exists social interactive behavior between Carl andJack on this publishedmessage it is probably that the emotionof Jackrsquos opinion is similar to Carlrsquos

From the social interactive relationship depicted in Fig-ure 1 we conceive that the support and opposition relationamong microblog participants can be exploited to deducethe sentiment of messages under the condition of pooremotion expressions Firstly it is necessary to build the socialinteractive relationship with mathematical formula Supposea message posted by user 119894 is supported by user 119895 that is tosay user 119895 keeps similar sentiment to user 119894 while if user 119894rsquosmessage is opposed by user 119895 it is regarded that user 119895 holdsopposite sentiment to user 119894 Via the interactive history wecan build an interactive relationshipmatrix119877 to represent theattribution based on the microblog messages

More specifically element 119903119894119895 in matrix 119877 denotes theinteractive relationship from user 119894 to user 119895 119903119894119895 = 1 meansuser 119894 has supported user 119895rsquos opinion in message while 119903

119894119895=

minus1 means user 119894 has opposed user 119895rsquos opinion hidden inthe message published by user 119895 If there exists no kindof this interactive relationship element 119903

119894119895is assigned to


0 1 0

1 0 0

1 0 0

0 1 0

0 0 1

P

UsersM

essa

ge

1 0 1

1 1 0

0 0 1

1 1 0

0 0 1

Mes

sage

Feature

0

1

0

0

1

F

0 1

0 1

1 0

R

Users

Use

rs

minus1

minus1

minus1

m1

m2

m3

m4

m5

m1

m2

m3

m4

m5

u1

u1

u2

u2

u3

u1 u2 u3

u3

f1 f2 f3 f4

(a)

m1

m1

m2

m2

m3

m3

m4

m4

m5

m5

0

1 0 0

1 0 0

0

1 1

0 1

1

1

0 1

0

M

minus1minus1

minus1minus1

minus1

minus1

minus1minus1

(b)

Figure 2 An example of microblog message matrix representation

1 The interactive relationship matrix 119877 for the above toyexample is listed as follows It is not hard to obtain thatCarlrsquos influence is powerful among the four people For otherpeople including Tom Jack and Jane all vote in favor of himThat is to say other peoplersquos sentiment polarity for certaintopic is homogenous to Carlrsquos It is noted that matrix 119877 is anonsymmetric matrix as the relationship between any twousers is bidirectional

119877 =

[

[

[

[

[

[

0 0 minus1 1

minus1 0 minus1 1

0 minus1 0 1

minus1 0 0 0

]

]

]

]

]

]

(1)

33 Sentiment Classification Model In the following weshould extract the interactive relation from posted messagesin data set Given a microblog message data set 119875 isin R119897times119899

can be used to represent the content matrix where 119897 is thenumber of messages and 119899 is the number of participant usersas shown in the left of Figure 2 119875

119894119895= 1 denotes that message

119898119894is published by user119906

119895 For eachmessage in the data set we

can build the message-feature vector And the correspondingmessage-feature matrix 119865 isin R119897times119902 where 119902 is the number ofextracted feature can be shown in Figure 2 And the user-usermatrix which is obtained from history messages is 119877 isin R119899times119899as shown in the right of Figure 2

Next we can define the problem of sentiment orientationidentification in this work Consider a microblog messagedata set 119875 isin R119897times119899 the message-feature representation forthe given message data set 119865 isin R119897times119902 and a user-user socialinteractive relationship matrix 119877 isin R119899times119899 and the goal ofsentiment analysis is to obtain classifier119882 to label sentimentautomatically

Inspired by the work [7] we propose a hybrid approachincorporating microblog social interactive relationship intotraditional content-based sentiment analysis process Specif-ically via modeling the social interactive relationship historywe establish a matrix to represent the voting relationshipamong many microblog users Afterwards we constructthe relationship for messages based on the usersrsquo socialinteractionThe relationship can be calculated by the formulaas follows

119872 = 119875 times 119877 times 119875119879 (2)

The elements inmatrix119872 representwhether twodifferentmessages are published by two participants inmicroblog withsocial interactive relationship If 119898119894119895 = 1 that means theauthor of message 119894 has a support interactive relationshipwith the author of message 119895 and sentiments of these twomicroblog messages are identical while 119898

119894119895= minus1 indicates

that message 119894rsquos holder has an opposed relationship withmessage 119895rsquos and the sentiments of the related two messagesare inverse As message 2 and 3 are all posted by user 1 in the


above example the values in the integrated matrix 119872 havesimilar polarity which are circled by red dotted lines

In this paper we employ the 119899-gram model to extractthe feature vector 119891119894 isin R119876times1 of textual messages where 119876is the dimension of feature vector By leveraging the LeastSquares method it is not hard to generate the content-basedsentiment classifier score More generally the classificationproblem can be transformed into an optimization problemas

min 1

2

10038171003817100381710038171003817119883119879119882minus 119884

10038171003817100381710038171003817

2

119865 (3)

where 119883 is the message to be analyzed 119882 denotes theclassifiers and 119884 is the sentiment label vector from context-based classifiersThe goal is to find a vector of119882 to make theFrobenius norm of matrix reaches minimum value

Moreover it is necessary to integrate the aforementionedsocial interactive relationship factor into the learning processdepicted in formula (2) The basic idea is to establish acorrelation to make two messages as similar as possible iftheir holders have a support interactive relationship andmaketwo messages different in emotion if their authors have anopposed relationship So the social interactive relationshipbetween messages can be formulated as follows

119873

sum

119894=1

119872119894119895

10038161003816100381610038161003816119871119894minus 119871119895

10038161003816100381610038161003816 (4)

where 119871 = 119883119879119882 119871 = 119871

1 1198712 1198713 119871

119873

So the sentiment classification can be solved by thesynthetical formula

1

2

119873

sum

119894=1

(119871119894minus 119884119894)2+ 120574

119873

sum

119894=1

119872119894119895

10038161003816100381610038161003816119871119894minus 119871119895

10038161003816100381610038161003816 (5)

4 Experiment and Result Analysis

Each experiment is repeated ten times independently and theaverage results are reported The experiments are conductedon microblog message data set which is crawled from SinaWeibo The data set consists of 37 million messages Thecorresponding social network is also included in the data setFirstly we conduct experiments to verify the performance ofour proposed SSTI (Sparse Sentiment Terms Identification)approach by comparing it with existing common sentimentclassification approaches including LS (Least Squares) andSVM (Support VectorMachine)The experimental results areshown in Figure 3 From there it is not hard to see that theperformance of proposed method SSTI is better than thatof the baseline methods on different data size The reason isthat the existing methods are purely content-based approachand ignore the social relationship Moreover the sentimentidentification accuracy for microblog messages increases asthe size of training data set increases

We also investigate parameter 120574 in our proposed senti-ment classification learned model The role of 120574 is to controlthe contribution of social interactive relationship factor in theoptimization formula Experiments are conducted to study

15 35 85 100The size of training data set ()

LSSVMSSTI

The i

dent

ifica

tion

accu

racy

03

04

05

06

07

08

09

02

1

Figure 3 Sentiment identification on microblog messages

0

01

02

03

04

05

06

07

08

09

1

The i

dent

ifica

tion

accu

racy

01 02 03 04 05 06 07 08 09 10The value of 120574

Figure 4 The influence of parameter 120574 in sentiment identification

the influence of 120574 in the sentiment identification processThe corresponding results are shown in Figure 4 as followsAs shown in Figure 4 when the value of parameter 120574 isnot large the performance of sentiment identification formicroblog messages could be improved as 120574 increases whileif the parameter increases to some extent the performancewill drop obviously The reason for this phenomenon is thatwhen the parameter achieves a large value the influence ofsocial interactive relationship factor is overstressed So it isnecessary to choose a suitable value for the parameter

Next we will discuss the guidelines for how to choosean appropriate value of parameter 120574 As explained abovethe value of parameter 120574 is to determine the weight ofsocial interactive relationship factor Although the socialinteraction can improve the sentiment classification accuracy


0

001

002

003

004

005

006

007

008

009

Runt

ime (

sec)

100 150 200 250 300 350 40050The data size

Figure 5 The runtime efficiency of our approach

of microblog message the information derived from contentitself is also important In other words the social interac-tive relationship and content semantic information shouldbe leveraged synchronously From Figure 4 the sentimentidentification accuracy can achieve best performance whenthe value of parameter 120574 varies from 04 to 06 That is tosay the parameter should be chosen from 04 to 06 For theexact value of parameter 120574 it may change slightly for differenttraining data set social culture and so on

Finally we explore the runtime efficiency of our proposedapproach The experiments are implemented with Interl(R)Core i3-3110M CPU 400GB RAM in Matlab R2010benvironment The runtime efficiency results are shown inFigure 5 From the experiment results in Figure 5 it is nothard to obtain that the runtime of our proposed approachis approximately linear with the size of data It takes 007seconds to finish the classification process when the datasize is equal to 400 The results validate the efficiency of ourapproach

5 Conclusions and Future Work

Sentiment classification of microblog message is an impor-tant research area Through classification and analysis ofsentiments on microblog one can get an understanding ofpeoplersquos attitudes about particular topics However some-times there are not enough emotion terms in the messagesto be analyzed The sparse sentiment terms in microblogmessage pose a challenge to the content-based sentimentclassification methods Towards this problem we proposea novel notion of social interactive relationship based onmicroblog messages and propose incorporating it in theexisting content-based approach After modeling the socialinteractive relationship matrix from user-message matrix webuild a sentiment identification learned model to analyzethe emotion of a given message Experiments demonstratethat our proposed approach can improve the sentimentclassification performance significantly

Conflict of Interests

The authors declare that there is no conflict of interestsregarding the publication of this paper

Acknowledgment

This work is supported by National Natural Science Founda-tion of China (Grant nos 61402360 and 61401015)

References

[1] MMMostafa ldquoMore than words social networksrsquo text miningfor consumer brand sentimentsrdquo Expert Systems with Applica-tions vol 40 no 10 pp 4241ndash4251 2013

[2] A Go R Bhayani and L Huang ldquoTwitter sentiment clas-sification using distant supervisionrdquo Project Report CS224NStanford University 2009

[3] M Ghiassi J Skinner andD Zimbra ldquoTwitter brand sentimentanalysis a hybrid system using n-gram analysis and dynamicartificial neural networkrdquo Expert Systems with Applications vol40 no 16 pp 6266ndash6282 2013

[4] A Kumar and T M Sebastian ldquoSentiment analysis on twitterrdquoInternational Journal of Computer Science Issues vol 9 no 3 pp372ndash378 2012

[5] H Saif Y He and H Alani ldquoSemantic sentiment analysis oftwitterrdquo inThe Semantic WebmdashISWC 2012 vol 7649 of LectureNotes in Computer Science pp 508ndash524 Springer BerlinGermany 2012

[6] J Bollen A Pepe and H Mao ldquoModelingpublic mood andemotion twitter sentiment and socio-economic phenomenardquohttparxivorgabs09111583

[7] X Hu L Tang J Tang and H Liu ldquoExploiting social relationsfor sentiment analysis in microbloggingrdquo in Proceedings of the6th ACM International Conference on Web Search and DataMining (WSDM rsquo13) pp 537ndash546 Rome Italy February 2013

[8] F Wu Y Huang and Y Song ldquoStructured microblog sentimentclassification via social context regularizationrdquo Neurocomput-ing vol 175 pp 599ndash609 2016

[9] L Jiang M Yu M Zhou X Liu and T Zhao ldquoTarget-dependent twitter sentiment classificationrdquo in Proceedings of the49th Annual Meeting of the Association for ComputationalLinguistics Human Language TechnologiesmdashVolume 1 pp 151ndash160 Association for Computational Linguistics June 2011

[10] B Pang and L Lee ldquoOpinion mining and sentiment analysisrdquoFoundations and Trends in Information Retrieval vol 2 no 1-2pp 1ndash135 2008

[11] A Bermingham and A F Smeaton ldquoClassifying sentimentin microblogs is brevity an advantagerdquo in Proceedings of the19th ACM International Conference on Information and Knowl-edge Management (CIKM rsquo10) pp 1833ndash1836 ACM TorontoCanada October 2010

[12] J Davis and S OrsquoFlaherty ldquoAssessing the accuracy of automatedtwitter sentiment codingrdquo Academy of Marketing Studies Jour-nal vol 16 supplement 1 p 35 2012

International Journal of

AerospaceEngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

RoboticsJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014


Active and Passive Electronic Components

Control Scienceand Engineering

Journal of



RotatingMachinery


Hindawi Publishing Corporation httpwwwhindawicom

Journal ofEngineeringVolume 2014

Submit your manuscripts athttpwwwhindawicom

VLSI Design



Shock and Vibration


Civil EngineeringAdvances in

Acoustics and VibrationAdvances in



Electrical and Computer Engineering

Journal of

Advances inOptoElectronics


Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

SensorsJournal of


Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014


Chemical EngineeringInternational Journal of Antennas and

Propagation




Navigation and Observation



DistributedSensor Networks



sparse sentiment terms problem [7 8] The research worksin [7 8] are based on the hypothesis that people with friendrelationship often keep similar sentiment towards a giventopic

Meanwhile in practice people may be in disagreementwith others in their friend circle as they have differentsocial cognition knowledge structure and experience Thatis to say one may oppose or accept the opinion reflectedin message posted by someone else And the oppositionor acceptation behavior also reflects the emotion polarityof the participants in the microblog environment In thispaper this opposition or acceptation behavior is regardedas a social interactive relationship By leveraging this kindof interactive relationship it may be feasible to infer thesentiment of message for participants and build the learningmodel In this work we exploit the social relationshipbetween users on microblog platform and incorporate thisfactor into content-based analysis process Via exploitingthe social interactive relationship among microblog users avoting-based social interactive model has been built to linkthe microblog participants based on the message contextAfterwards the sentiment classification problem can betransformed into an optimization problem integrating thecontent-based strategy and social interactive relationshipfactor Experimental results on microblog message data setsvalidate the effectiveness and efficiency of our method

This paper is structured as follows In Section 2 wereview related work on sentiment classification After that wedescribe and formulate the social interactive relationship inmicroblog message and propose the classification model indetail in Section 3 In Section 4 we then evaluate our modelsand present experimental results Finally in Section 5 wesummarize our findings and contributions

2 Background and Related Works

In recent years as an opinion-rich resource microblogmessages provide the possibility of mining and analyzing theemotion of people in different scales And microblog senti-ment orientation identification has gained hung popularityand attracted researchersrsquo attention across many fields In thissection an overview of works related to microblog sentimentanalysis and classification is presented

Previous works usually compare the text content of agivenmessagewith a lexicon or a dictionary to classify its sen-timent and calculate its strength For instance SentiWordNetcontains as many as 200000 entries to match each word withpositive negative or objective scores Mostafa et al proposea lexicon-based method to analyzing consumersrsquo sentimenttowards some commercial brands [1 2] Go et al [2] studythe problem of sentiment classification for Twitter messagesby the means of machine learning algorithms Based ondistant supervised learning thework compares the sentimentclassification performance for three algorithms includingNaıve Bayes Maximum Entropy and SVM Ghiassi et al [3]introduce an approach to supervised feature reduction using119899-grams and statistical analysis to develop a Twitter-specificlexicon for sentiment analysis Kumar and Sebastian [4]

expound a hybrid approach to analyze the sentiment of Twit-terThe approach employs a corpus-basedmethod to identifythe semantic orientation of adjectives and a dictionary-basedmethod to find the semantic orientation of verbs and adverbsSaif et al [5] add semantics information as additional featuresin the process of tweets sentiment analysis Moreover thecorrelation of the representative concept is calculated withpositive and negative sentiment Bollen et al [6] extract sixdimensions of emotion using an established psychometricinstrument POMS and preform tweet sentiment analysis inthe field of stock market crude oil price indices and majorpublic events

As tweets are usually short and more ambiguous Jianget al [9] study the target-dependent Twitter sentiment clas-sification problem Through incorporating target-dependentfeatures and context information related to the given targettweet the sentiment of query tweet can be classified aspositive negative or neutral Pang and Lee [10] review therelated approaches for opinion mining and sentiment analy-sis Besides some other issues including privacy protectionmanipulation and economic impart are also discussed in[10] Althoughmany researchers point out that the propertiesof short and noisy present new challenges to the sentimentanalysis for microblog message Bermingham and Smeaton[11] find classifying sentiment in microblogs easier than inblogs Davis and OrsquoFlaherty [12] evaluate the automatedsentiment coding accuracy and misclassification errors of sixleading third-party companies for a broad range of commenttypes and forms

Actually beside the content-based information the hid-den social relationship in microblog message can also bereflected and may be used in the sentiment analysis processRecently there are many works that focus on the socialrelationship in microblog space and leveraged the relation-ship to improve the classification accuracy [7 8] Hu etal [7] exploit social relationship and introduce two socialtheory hypotheses into social network which are sentimentconsistency and emotional contagion Pang and Lee [10]expand the hypothesis verified by Hu et al in [7] andproposed a microblog sentiment classification frameworkInspired by the work in [7] we consider combining theimplicit social relationship behind the microblog message toclassify its sentiment polarity

3 Problem Definition

31 Motivation The purpose of sentiment orientation iden-tification for microblogs is to build a program which canautomatically identify whether a given microblog messageis expressing positive negative or no sentiment In otherwords this problem can be defined as follows given acollection of microblog message set 119863 and a set of classes119875119866 = positive neutralnegative the goal is to constructa mapping function Φ119866 119863 rarr 119875119866 which describes thesentiment orientation of messages according to an integratedmodel

However the character of microblog message such asnoisy short and ambiguous poses a challenge for sentimentanalysis and mining In addition the informal language








Carl

Jack

Tom

Jane

AgreeOppose






119894119895=




0 1 0

1 0 0

1 0 0

0 1 0

0 0 1

P

UsersM

essa

ge

1 0 1

1 1 0

0 0 1

1 1 0

0 0 1

Mes

sage

Feature

0

1

0

0

1

F

0 1

0 1

1 0

R

Users

Use

rs

minus1

minus1

minus1

m1

m2

m3

m4

m5

m1

m2

m3

m4

m5

u1

u1

u2

u2

u3

u1 u2 u3

u3

f1 f2 f3 f4

(a)

m1

m1

m2

m2

m3

m3

m4

m4

m5

m5

0

1 0 0

1 0 0

0

1 1

0 1

1

1

0 1

0

M

minus1minus1

minus1minus1

minus1

minus1

minus1minus1

(b)



119877 =

[

[

[

[

[

[

0 0 minus1 1

minus1 0 minus1 1

0 minus1 0 1

minus1 0 0 0

]

]

]

]

]

]

(1)









119872 = 119875 times 119877 times 119875119879 (2)







min 1

2

10038171003817100381710038171003817119883119879119882minus 119884

10038171003817100381710038171003817

2

119865 (3)



119873

sum

119894=1

119872119894119895

10038161003816100381610038161003816119871119894minus 119871119895

10038161003816100381610038161003816 (4)

where 119871 = 119883119879119882 119871 = 119871

1 1198712 1198713 119871

119873


1

2

119873

sum

119894=1

(119871119894minus 119884119894)2+ 120574

119873

sum

119894=1

119872119894119895

10038161003816100381610038161003816119871119894minus 119871119895

10038161003816100381610038161003816 (5)





LSSVMSSTI

The i

dent

ifica

tion

accu

racy

03

04

05

06

07

08

09

02

1


0

01

02

03

04

05

06

07

08

09

1

The i

dent

ifica

tion

accu

racy

01 02 03 04 05 06 07 08 09 10The value of 120574





0

001

002

003

004

005

006

007

008

009

Runt

ime (

sec)

100 150 200 250 300 350 40050The data size








Acknowledgment


References















RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation
















Carl

Jack

Tom

Jane

AgreeOppose






119894119895=




0 1 0

1 0 0

1 0 0

0 1 0

0 0 1

P

UsersM

essa

ge

1 0 1

1 1 0

0 0 1

1 1 0

0 0 1

Mes

sage

Feature

0

1

0

0

1

F

0 1

0 1

1 0

R

Users

Use

rs

minus1

minus1

minus1

m1

m2

m3

m4

m5

m1

m2

m3

m4

m5

u1

u1

u2

u2

u3

u1 u2 u3

u3

f1 f2 f3 f4

(a)

m1

m1

m2

m2

m3

m3

m4

m4

m5

m5

0

1 0 0

1 0 0

0

1 1

0 1

1

1

0 1

0

M

minus1minus1

minus1minus1

minus1

minus1

minus1minus1

(b)



119877 =

[

[

[

[

[

[

0 0 minus1 1

minus1 0 minus1 1

0 minus1 0 1

minus1 0 0 0

]

]

]

]

]

]

(1)









119872 = 119875 times 119877 times 119875119879 (2)







min 1

2

10038171003817100381710038171003817119883119879119882minus 119884

10038171003817100381710038171003817

2

119865 (3)



119873

sum

119894=1

119872119894119895

10038161003816100381610038161003816119871119894minus 119871119895

10038161003816100381610038161003816 (4)

where 119871 = 119883119879119882 119871 = 119871

1 1198712 1198713 119871

119873


1

2

119873

sum

119894=1

(119871119894minus 119884119894)2+ 120574

119873

sum

119894=1

119872119894119895

10038161003816100381610038161003816119871119894minus 119871119895

10038161003816100381610038161003816 (5)





LSSVMSSTI

The i

dent

ifica

tion

accu

racy

03

04

05

06

07

08

09

02

1


0

01

02

03

04

05

06

07

08

09

1

The i

dent

ifica

tion

accu

racy

01 02 03 04 05 06 07 08 09 10The value of 120574





0

001

002

003

004

005

006

007

008

009

Runt

ime (

sec)

100 150 200 250 300 350 40050The data size








Acknowledgment


References















RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation










0 1 0

1 0 0

1 0 0

0 1 0

0 0 1

P

UsersM

essa

ge

1 0 1

1 1 0

0 0 1

1 1 0

0 0 1

Mes

sage

Feature

0

1

0

0

1

F

0 1

0 1

1 0

R

Users

Use

rs

minus1

minus1

minus1

m1

m2

m3

m4

m5

m1

m2

m3

m4

m5

u1

u1

u2

u2

u3

u1 u2 u3

u3

f1 f2 f3 f4

(a)

m1

m1

m2

m2

m3

m3

m4

m4

m5

m5

0

1 0 0

1 0 0

0

1 1

0 1

1

1

0 1

0

M

minus1minus1

minus1minus1

minus1

minus1

minus1minus1

(b)



119877 =

[

[

[

[

[

[

0 0 minus1 1

minus1 0 minus1 1

0 minus1 0 1

minus1 0 0 0

]

]

]

]

]

]

(1)









119872 = 119875 times 119877 times 119875119879 (2)







min 1

2

10038171003817100381710038171003817119883119879119882minus 119884

10038171003817100381710038171003817

2

119865 (3)



119873

sum

119894=1

119872119894119895

10038161003816100381610038161003816119871119894minus 119871119895

10038161003816100381610038161003816 (4)

where 119871 = 119883119879119882 119871 = 119871

1 1198712 1198713 119871

119873


1

2

119873

sum

119894=1

(119871119894minus 119884119894)2+ 120574

119873

sum

119894=1

119872119894119895

10038161003816100381610038161003816119871119894minus 119871119895

10038161003816100381610038161003816 (5)





LSSVMSSTI

The i

dent

ifica

tion

accu

racy

03

04

05

06

07

08

09

02

1


0

01

02

03

04

05

06

07

08

09

1

The i

dent

ifica

tion

accu

racy

01 02 03 04 05 06 07 08 09 10The value of 120574





0

001

002

003

004

005

006

007

008

009

Runt

ime (

sec)

100 150 200 250 300 350 40050The data size








Acknowledgment


References















RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation












min 1

2

10038171003817100381710038171003817119883119879119882minus 119884

10038171003817100381710038171003817

2

119865 (3)



119873

sum

119894=1

119872119894119895

10038161003816100381610038161003816119871119894minus 119871119895

10038161003816100381610038161003816 (4)

where 119871 = 119883119879119882 119871 = 119871

1 1198712 1198713 119871

119873


1

2

119873

sum

119894=1

(119871119894minus 119884119894)2+ 120574

119873

sum

119894=1

119872119894119895

10038161003816100381610038161003816119871119894minus 119871119895

10038161003816100381610038161003816 (5)





LSSVMSSTI

The i

dent

ifica

tion

accu

racy

03

04

05

06

07

08

09

02

1


0

01

02

03

04

05

06

07

08

09

1

The i

dent

ifica

tion

accu

racy

01 02 03 04 05 06 07 08 09 10The value of 120574





0

001

002

003

004

005

006

007

008

009

Runt

ime (

sec)

100 150 200 250 300 350 40050The data size








Acknowledgment


References















RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation










0

001

002

003

004

005

006

007

008

009

Runt

ime (

sec)

100 150 200 250 300 350 40050The data size








Acknowledgment


References















RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation











RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation









Documents

Research Article Microblog Sentiment Orientation …downloads.hindawi.com/journals/jece/2016/7282913.pdfclassify its sentiment polarity. 3. Problem Definition. . Motivation. e purpose