ࡱ> 352bjbjUU v(??$ %%%%%%; GXO%%OO%%OOOO%%OOOOO%3H^OO0OOOOHOOOOOOOOOOOOOOOOOOOOOOOOO :Resumo Tweets, no cenrio eleitoral, tm sido utilizados em pesquisas cientficas sob diversas perspectivas, por exemplo para predizer o resultado de eleies presidenciais e para investigar reaes de eleitores durante eventos, como debates eleitorais. A minerao de opinies tem sido uma das abordagens utilizadas para avaliar a opinio expressa por usurios nesse tipo de mensagem. Melhorar a acurcia do sentimento dessa categoria de tweet tem sido um dos desafios enfrentados pelos pesquisadores, devido alguns fatores, tais como a quantidade limitada de caracteres permitidos em um tweet e o uso de hashtags e slogans de campanha para expressar opinies polticas sobre candidatos. No cenrio de eleies, hashtags tm sido utilizadas em tweets com uma frequncia cada vez maior. Em diversos trabalhos reportados na literatura, hashtags tm sido utilizadas para coletar e selecionar tweets, rotular mensagens, alm de receberem tratamentos especiais na fase de pr-processamento ou so simplesmente descartadas das anlises. A contribuio de hashtags na anlise de sentimentos de tweets no cenrio eleitoral tem sido pouco explorada. Neste trabalho, so consideradas duas categorias de hashtags, as polticas e as no-polticas, e dois novos atributos baseados em hashtags polticas, denominados TPSB e DPSB, que refletem na melhoria do desempenho de algoritmos de aprendizado de mquina supervisionado, utilizados no processo de classificao do sentimento de tweets no cenrio eleitoral. A anlise experimental proposta neste trabalho, foi realizada a partir de duas amostras rotuladas manualmente contendo mensagens coletadas em perodos de eleies presidenciais, cada uma com, aproximadamente, 4.000 tweets. Foi avaliada a contribuio do uso das hashtags contidas em tweets e em descries de perfis de usurios, e atributos para a melhoria da acurcia de classificadores baseados em Naive Bayes (NB), Multinomial Naive Bayes (MNB) e Support Vector Machine (SVM). Nos testes realizados com uma das amostras, as hashtags polticas foram responsveis por aumentar, com significncia estatstica, as acurcias de todos os classificadores. Em outro dataset, as acurcias de todos os classificadores foram incrementadas e, em 66% dos casos, os aumentos obtidos foram estatisticamente significantes. Em um dos experimentos propostos, a acurcia do algoritmo NB, utilizando o formato unigrama, foi incrementada em 3,2% ao considerar hashtags polticas nas anlises. Os resultados obtidos sugerem que hashtags contendo palavras fazendo referncia a candidatos, a partir do primeiro nome e/ou sobrenome deles associado a outras palavras/nmeros, slogans de campanha e palavras de repdio, podem ser teis na classificao do sentimento de tweets polticos e que usurios do Twitter, durante perodos de campanha eleitoral, expressam opinio poltica no somente a partir de tweets, mas tambm a partir de suas descries de perfis. Palavras-chave: twitter, tweet, hashtag, descrio do perfil do usurio, poltica, minerao de opinio, anlise de sentimentos. Abstract Tweets in the election scene have been used in scientific research from a variety of perspectives, for example to predict the outcome of presidential elections and to investigate voter reactions during events such as electoral debates. The opinion mining has been one of the approaches used to evaluate the opinion expressed by the user in this type of message. Improving the accuracy of this category of tweet has been one of the challenges faced by researchers due to factors such as the limited amount of characters allowed in a tweet and the use of hashtags and campaign slogans to express political opinions about candidates. In the election scene, hashtags have been used in tweets with increasing frequency. In several papers reported in the literature, hashtags have been used to collect and select tweets, label messages, and receive special treatments in the preprocessing phase or are simply discarded from the analyzes. The contribution of hashtags in the sentiment tweets analysis in the election scene has been little explored. In this work, two categories of hashtags, the policies and the non-policies, and two new features based on political hashtags, called TPSB and DPSB, are considered, which reflect on the performance improvement of supervised machine learning algorithms used in the tweets sentiment classification in the election scene. The experimental analysis proposed in this work was performed from two manually labeled samples containing messages collected during periods of presidential elections, each with approximately 4,000 tweets. The contribution of the use of hashtags contained in tweets and descriptions of user profiles and attributes to improve the accuracy of Naive Bayes (NB), Multinomial Naive Bayes (MNB) and Support Vector Machine (SVM) classifiers was evaluated. In the tests performed with one of the samples, political hashtags were responsible for increasing, with statistical significance, the accuracy of all classi_ers. In another dataset, the accuracy of all classifiers was increased and, in 66% of cases, the increases obtained were statistically significant. In one of the proposed experiments, the accuracy of the NB algorithm, using the unigram format, was increased by 3.2% when considering political hashtags in the analyzes. The results obtained suggest that hashtags containing words referring to candidates, from their first name and/or surname associated with other words/numbers, campaign slogans and repudiation words, may be useful in classifying the political tweets sentiment and that Twitter users, during campaigning periods, express political opinion not only from tweets but also from their profile descriptions. Keywords: twitter, tweet, hashtag, description of user profile, politics, opinion mining, sentiment analysis.  N S T _ g h j q r D L M {  # $ A G H      $ ͹ͫͫͫͫͫͫͫͫͫͫͫͫhCJOJQJ^JaJ&hh6CJOJQJ]^JaJhCJOJQJ^JaJ&hh6CJOJQJ]^JaJhCJOJQJ^JaJ hmhCJOJQJ^JaJ&hmh5CJOJQJ\^JaJ6{|k$d7$8$H$a$gdmd7$8$H$gdm$d7$8$H$a$gdm$d7$8$H$a$gd`.d7$8$H$gd,$d7$8$H$a$gd, kvwxz}pw?A\cd@F|~ hmhCJOJQJ^JaJ h5CJOJQJ\^JaJhCJOJQJ^JaJ&hmh6CJOJQJ]^JaJhCJOJQJ^JaJhCJOJQJ^JaJ&hh6CJOJQJ]^JaJhCJOJQJ^JaJ1 h,hhCJOJQJ^JaJ&h*aWh5CJOJQJ\^JaJ6P1h. A!"#$% Dp<P1h:pC. A!"#$% Dpj  666666666666666666666666666666666666666666 6666666666 666666666666 6666666666666666666666666666666666666666666666666666666666666666662 0@P`p2( 0@P`p 0@P`p 0@P`p 0@P`p 0@P`p 0@P`p8XV~ OJPJQJ_HmHnHsHtHN`N CNormal dCJ^J_HaJmHsHtH DA D 0Default Paragraph FontRiR 0 Table Normal4 l4a (k ( 0No List PK![Content_Types].xmlj0Eжr(΢Iw},-j4 wP-t#bΙ{UTU^hd}㨫)*1P' ^W0)T9<l#$yi};~@(Hu* Dנz/0ǰ $ X3aZ,D0j~3߶b~i>3\`?/[G\!-Rk.sԻ..a濭?PK!֧6 _rels/.relsj0 }Q%v/C/}(h"O = C?hv=Ʌ%[xp{۵_Pѣ<1H0ORBdJE4b$q_6LR7`0̞O,En7Lib/SeеPK!kytheme/theme/themeManager.xml M @}w7c(EbˮCAǠҟ7՛K Y, e.|,H,lxɴIsQ}#Ր ֵ+!,^$j=GW)E+& 8PK!Ptheme/theme/theme1.xmlYOo6w toc'vuر-MniP@I}úama[إ4:lЯGRX^6؊>$ !)O^rC$y@/yH*񄴽)޵߻UDb`}"qۋJחX^)I`nEp)liV[]1M<OP6r=zgbIguSebORD۫qu gZo~ٺlAplxpT0+[}`jzAV2Fi@qv֬5\|ʜ̭NleXdsjcs7f W+Ն7`g ȘJj|h(KD- dXiJ؇(x$( :;˹! I_TS 1?E??ZBΪmU/?~xY'y5g&΋/ɋ>GMGeD3Vq%'#q$8K)fw9:ĵ x}rxwr:\TZaG*y8IjbRc|XŻǿI u3KGnD1NIBs RuK>V.EL+M2#'fi ~V vl{u8zH *:(W☕ ~JTe\O*tHGHY}KNP*ݾ˦TѼ9/#A7qZ$*c?qUnwN%Oi4 =3ڗP 1Pm \\9Mؓ2aD];Yt\[x]}Wr|]g- eW )6-rCSj id DЇAΜIqbJ#x꺃 6k#ASh&ʌt(Q%p%m&]caSl=X\P1Mh9MVdDAaVB[݈fJíP|8 քAV^f Hn- "d>znNJ ة>b&2vKyϼD:,AGm\nziÙ.uχYC6OMf3or$5NHT[XF64T,ќM0E)`#5XY`פ;%1U٥m;R>QD DcpU'&LE/pm%]8firS4d 7y\`JnίI R3U~7+׸#m qBiDi*L69mY&iHE=(K&N!V.KeLDĕ{D vEꦚdeNƟe(MN9ߜR6&3(a/DUz<{ˊYȳV)9Z[4^n5!J?Q3eBoCM m<.vpIYfZY_p[=al-Y}Nc͙ŋ4vfavl'SA8|*u{-ߟ0%M07%<ҍPK! ѐ'theme/theme/_rels/themeManager.xml.relsM 0wooӺ&݈Э5 6?$Q ,.aic21h:qm@RN;d`o7gK(M&$R(.1r'JЊT8V"AȻHu}|$b{P8g/]QAsم(#L[PK-![Content_Types].xmlPK-!֧6 +_rels/.relsPK-!kytheme/theme/themeManager.xmlPK-!Ptheme/theme/theme1.xmlPK-! ѐ' theme/theme/_rels/themeManager.xml.relsPK] (8(_GoBack = q-`.; 0Y=C*aWk,m@@@Unknown G*Ax Times New Roman5Symbol3" Arial/ F43/ F165CMMI12/ F487.@CalibriACambria Math"tGtGa0$P,!xxResumo Wilton FilhoHelioOh+'0  < H T`hpxResumoWilton FilhoNormal_WordconvHelio2Microsoft Office Outlook@@}-^@}-^a՜.+,0 hp|   Resumo Title  !#$%&'()+,-./014Root Entry FH^61TableWordDocumentv(SummaryInformation("DocumentSummaryInformation8*CompObjy  F'Microsoft Office Word 97-2003 Document MSWordDocWord.Document.89q  F#Documento do Microsoft Office Word MSWordDocWord.Document.89q