TweeProfiles3: Visualização de Padrões Espácio-Temporais no Twiter

FACULDADE DE ENGENHARIA DA UNIVERSIDADE DO PORTO TweeP o iles3: isualização de pad ões espacio- empo ais no Twi e And é Filipe do Cou o Maia Mes ado In eg ado em Engenha ia Ele o écnica e de Compu ado es O ien ado : P o . Ca los Soa es Co-O ien ado : P o . Ped o Ab eu Janua y 26, 2015 c And é Filipe do Cou o Maia, 2015 i ii Resumo Com o ad en o das edes sociais, uma g ande quan idade de dados do u ilizado oi ge ada. Desde há alguns anos, os in es igado es e as emp esas ape cebe am-se do alo associado a es as eno mes quan idades de dados e dando o igem ao desen ol imen o de algo i mos e e amen as pa a ex ai pad ões e pa a os usa . O TweeP o iles é uma e amen a de clus e ing que pe mi e analisa os dados dos wee s sob múl iplas dimensões: espacial, empo al, con eúdo e social. O obje i o oi o desen ol imen o de uma aplicação web pa a iden i icação de pad ões em mensagens no Twi e . A é ago a, o am de- sen ol idas duas ex ensões pa a es e p oje o, TweeP o iles2 e Olhó-Passa inho. Ambos ealizam ag upamen o sob e os dados do Twi e , mas com ca ac e ís icas únicas: o TweeP o iles2, p ocessa dados em empo eal, e Olhó-Passa inho que ac escen ou à dimensão de con eúdo as imagens. Todo o abalho ealizado no TweeP o iles a é es a disse ação oi essencialmen e écnico e cien í- ico com o obje i o de concebe e p o o ipa soluções pa a os desa ios en ol idos. Po es a azão, ainda não oi possí el aze uma a aliação da e amen a numa aplicação do mundo eal. O ob- jec i o des a disse ação é da um p imei o passo nes e sen ido, em que o domínio de aplicação é o jo nalismo. Pa a a ingi es e objec i o, começámos po aze um le an amen o dos equisi os especí icos desse domínio de aplicação. Com base nesses equisi os e numa a aliação do es ado do TweeP o iles2, o am ealizadas algumas a e as de manu enção e desen ol imen o pa a consolida a e amen a. Pa a além da esolução de alguns p oblemas na implemen ação exis en e, es a consolidação e e po obje i os melho a an o a ecolha de dados como a in e ação com o u ilizado , aspe os essen- ciais pa a pode mos passa o sis ema pa a p odução. O mecanismo de ecolha de dados oi subs i- uido pelo SocialBus. O SocialBus é uma e amen a de ecolha, p ocessamen o e a mazenamen o de dados de edes sociais, em pa icula do Twi e . Em elação à in e ação com o u ilizado , oi e- alizado um inqué i o com pessoas expe ien es em jo nalismo, a im de en ende as necessidades e desejos de uma pla a o ma como o TweeP o iles3. Foi implemen ado um p ocesso de isualização adequado pa a o luxo de dados, usando á ios widge s pa a melho ep esen a oda a in o maçao. O sis ema oi desen ol ido endo em con a o es ado da a e dos p oje os nes a á ea e os esul ados ob idos do es udo do u ilizado , pa a pode mos da um passo em en e. A u ilidade da e amen a desen ol ida pa a o jo nalismo oi a aliada com base num es e de us- abilidade. Apesa de e sido ealizado com um pequeno conjun o de u ilizado es, es e es e se iu pa a a ingi os obje i os do p oje o, nomeadamen e, aze uma p imei a a aliação da u ilidade da e amen a TweeP o iles numa aplicação eal. Os esul ados ob idos pe mi i am não só as suas po encialidades como ques ões a melho a . iii i Abs ac Wi h he ad en o social ne wo king, a lo o use -speci ic, olun a ily p o ided da a has been gene a ed. A ew yea s ago, esea che s and companies no iced he alue ha lied wi hin hose eno mous amoun s o da a and de eloped algo i hms and ools o ex ac pa e ns om hose da a and o use hem. TweeP o iles is a clus e ing ool ha analyses wee s o e mul iple dimensions: spa ial, empo al, con en and social. The goal was o de elop a web applica ion o iden i y pa e ns in Twi e pos s. So a , he e ha e been wo ex ensions o his p ojec , TweeP o iles2 and Olhó-Passa inho. Bo h pe o m clus e ing o e Twi e da a bu wi h unique ea u es: TweeP o iles2 p ocesses eal- ime da a and Olhó-Passa inho in eg a ed he analysis o images in he con en dimension. All wo k in TweeP o iles, up o his disse a ion, was essen ially echnical and scien i ic in o de o design and p o o ype solu ions o he challenges iden i ied. Fo his eason, i has no ye been possible o e alua e he ool in a eal-wo ld applica ion. The aim o his wo k is a i s s ep in his di ec ion, in which he applica ion domain is jou nalism. To achie e his goal, we s a ed by making a su ey o he speci ic equi emen s o his appli- ca ion domain. Based on hese equi emen s and an assessmen o he s a e o TweeP o iles2 some main enance and de elopmen ac i i ies we e ca ied ou o consolida e i . In addi ion o sol ing some p oblems, his consolida ion aimed o imp o e bo h he da a collec ion p ocess as well as use in e ac ion, essen ial aspec s in o de o swi ch he sys em o p oduc ion. The da a collec ion mechanism has been eplaced by he SocialBus pla o m. SocialBus is a ool o he collec ion, p ocessing and s o age o da a om social ne wo ks, namely Twi e . Rega ding he in e ac ion wi h he use , an in es iga ion wi h a small g oup o expe ienced people in jou nalism we e su eyed, in o de o unde s and he needs and desi es o a pla o m such as TweeP o iles3. A isualiza ion p ocess sui able o da a s eaming was designed, using mul iple widge s o be e ep esen all he in o ma ion. The sys em was de eloped aking in o accoun he s a e o he a p ojec s in his a ea and he esul s ob ained om he use s udy, in o de o mo e he ool one s ep o wa d. The use ulness o he de eloped ool o jou nalism was e alua ed based on a usabili y es . Al- hough i was ca ied ou wi h a small se o use s, i was su icien o achie e he objec i es o he p ojec . In pa icula i enabled a i s assessmen o he usage o TweeP o iles in a eal applica ion, iden i ying no only i s po en ial as well as issues ha need imp o emen .. i Ag adecimen os Começo po ag adece ao meu o ien ado , Ca los Soa es, pela opo unidade de ealiza es e p o- jec o em colabo ação com o SAPO, pela o ien ação global do abalho e p opos as de soluções du an e odo o desen ol imen o. Ag adeço igualmen e ao meu co-o ien ado , Ped o Ab eu, po oda a ajuda du an e odo o p o- jec o. Sem es as duas pessoas, es a disse ação não e ia ido pe nas pa a anda . Ag adeço ambém ao (D .) Tiago Cunha. Um since o ob igado po oda a ajuda, disponibilidade e apoio o necido du an e es es úl imos meses. Uma pala a de ap eço a odos os que me acompanha am du an e es a (longa) jo nada. A odos aqueles, que de capa e ba ina, aziam um ano no mal, num ano melho , pa a odos nós, c iando memó ias que nos ão acompanha pa a semp e e amizades que não se ão pe de . Que o ambém ag adece ao SAPO Labs po oda a disponibilidade pa a ajuda no p ojec o. Um ob igado aos meus companhei o do Pub 1808, po semp e se in e essa em na disse ação e apoia em em odos os momen os. Po úl imo, mas com mais impo ância, que o ag adece à minha amílias, especialmen e aos meus pais, po odo o apoio que me de am e po ac edi a em semp e no meu sucesso, mesmo quando as coisas es a am mais complicadas. À magi, Paula Viei a, um ag adecimen o (ainda mais) especial, que sem ela, es a jo nada nunca e ia sido a mesma. Foi (e é) a minha maio companhia e o meu maio apoio, em odos os bons ou maus momen os, a minha maio o ça pa a nunca desis i . And é Maia ii xi LIST OF FIGURES 3.13 Example o en i ies in he es ing da ase . . . . . . . . . . . . . . . . . . . . . . 42 3.14 Example o news ob ained om he lis o en i ies. . . . . . . . . . . . . . . . . 42 4.1 Volume and spa ial dis ibu ion o wee s in he es ing da ase . . . . . . . . . . . 43 4.2 Posi ion o he wee s in he es ing da ase . . . . . . . . . . . . . . . . . . . . . 43 4.3 Spa ial Clus e ing esul s. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45 4.4 Tempo al Clus e ing esul s. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46 4.5 Tempo al Clus e ing esul s. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46 4.6 Tempo al and Spa ial Clus e ing esul s. . . . . . . . . . . . . . . . . . . . . . . 47 4.7 Con en Clus e ing esul s. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48 4.8 Con en Clus e ing esul s. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48 4.9 Con en Clus e ing esul s. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49 4.10 Con en and Spa ial Clus e ing esul s. . . . . . . . . . . . . . . . . . . . . . . . 49 4.11 TweeP o iles spa ial clus e ing esul s. . . . . . . . . . . . . . . . . . . . . . . . 50 4.12 TweeP o iles empo al clus e ing esul s. . . . . . . . . . . . . . . . . . . . . . . 50 4.13 Su ey ques ions ega ding ea u es om TweeP o iles3. . . . . . . . . . . . . . 52 A.1 TweeP o iles3mockup................................ 64 B.1 Ini ialGan Diag am................................. 68 B.2 FinalGan Diag am. ................................ 69 Lis o Tables 2.1 Analysis o di e en sys ems wi h Twi e da a clus e ing . . . . . . . . . . . . . 9 2.2 Di e ences be ween Visualiza ion Sys ems . . . . . . . . . . . . . . . . . . . . 16 2.3 Di e ences be ween TweeP o iles, TweeP o iles2 and Olhó-Passa inho . . . . . 24 2.4 Ex e nal lib a ies used in TweeP o iles2 [10].................... 26 4.1 Se o es spe o med. ............................... 44 4.2 Hyb idDenS eam pa ame e s. . . . . . . . . . . . . . . . . . . . . . . . . . . . 44 x x i LIST OF TABLES Symbols and Abb e ia ions 3D Th ee dimensional 2D Two dimensional DB Da abase AI A i icial In elligence API Applica ion P og amming In e ace DBSCAN Densi y-Based Spa ial Clus e ing o Applica ions wi h Noise JPN Jo nalismo Po o Ne x ii Chap e 1 In oduc ion Social ne wo ks ha e a majo impac nowadays. Thei in luence is el in se e al aspec s. Online social ne wo ks p esen a a ie y o social media se ices which ha e achie ed a huge impo ance in social li e as well as in ma ke ing s a egies as hey "ha e been ega ded as a imely and cos - e ec i e sou ce o spa io- empo al in o ma ion" [4]. One business which has been signi ican ly a ec ed is jou nalism. Social ne wo ks can be used o change he way jou nalis s a e able o ake he pulse o he ending hemes o opics ha a e cu en ly being alked abou . Recen ly, i acqui ed a new meaning in in o ma ion science which is "a dedica ed websi e o o he applica ion which enables use s o communica e wi h each o he " [11]. The massi e adhesion and he numbe o pla o ms ha p o ide social in e ac ion lead o a g ow h in he da a s o ed wi hin hese se ices. This da a has been used by many in es iga o s as a sou ce o in o ma ion [12,13,14]. Unlike wha happens wi h o he social ne wo ks like Facebook1and Linkedin2 ha use a ne - wo k o bi-di ec ional communica ion, Twi e uses an asymme ic in as uc u e whe e he e a e " iends" and " ollowe s". Assuming ha you a e a use o Twi e , he " iends" co espond o he accoun s o people ha you ollow and " ollowe s" co espond o he accoun s o people who ollow you [15]. Twi e is one o he op social ne wo ks, bo h in popula i y (wo ldwide pub- lic awa eness) and mon hly ac i e use s (a ound 250 million [16]). I is a s a ing poin o ou ongoing wo k because i is he mos widely used mic oblogging applica ion, wi h he numbe o ’ wee s’ eaching 110 million pe day in Janua y 2011 and is s ill escala ing [7]. The ini ial TweeP o iles [17,18] is ocused on iden i ying p o iles on da a collec ed [in o ma ion and ex ac ion ha e speci ic meanings in his con ex which may be misleading] om Twi e . The da a is p ocessed o e 4 dimensions (spa ial, empo al, social and con en ) using Da a Mining echniques. The ool enables he isualiza ion o he esul s o he clus e ing algo i hm. Since i lacked he abili y o p oduce eal- ime isualiza ions o he e olu ion o he da a s eam, as well as he abili y ind pa e ns in he images a ached o wee s, wo ex ensions we e de eloped: 1Mo e in o ma ion a : h ps:// acebook.com 2Mo e in o ma ion a : h ps://linkedin.com 1 2In oduc ion TweeP o iles2 [10] and Olhó-Passa inho [19]. Bo h ha e he same goal as TweeP o iles, o iden- i y p o iles on mul iple dimensions. Due o an unsa is ac o y handling o he social dimension in he o iginal ool, hese ex ensions ins ead o p ocessing he da a o e ou dimensions, only use h ee: spa ial, empo al and con en . TweeP o iles2 eplaced he o iginal ba ch clus e ing algo- i hm wi h a s eam clus e ing algo i hm, enabling he use o eal- ime da a. Howe e , he sys em was no ully implemen ed and s ill used s a ic da a o pe o m clus e ing, and he esul s ob ained we e as good as hey we e p e iously in TweeP o iles. Olhó-Passa inho in oduced a new se o esul s, since i was he i s TweeP o iles e sion using images as pa o he con en o wee s. Un il now, all wo k done was essen ially echnical and scien i ic, in o de o design and p o o- ype solu ions o he challenges in ol ed. Fo his eason, i has no ye been possible o e alua e he isualiza ion pla o m in a eal-wo ld applica ion.. Visualiza ion en i onmen s and echniques p o ide an impo an unc ion in communica ing u ban esea ch and suppo collabo a i e endea - ou s [20]. Big ad ances in digi al echnologies and he wide a ailabili y o he In e ne enables p oducing, manipula ing, and sha ing as digi al da a esou ces, many o which con ain geospa- ial e e ences [21]. Wi h mo e digi al da a becoming inc easingly a ailable he e a e no el ways u ban esea che s can now explo e u ban space and place in no el ways, suppo ed by a wide a ay o isualiza ion ools and echniques. Some o hese ools use Da a Mining echniques o gene a e knowledge. Some o he main unc ionali ies al eady implemen ed in his pla o m a e he isualiza ion o he loca ion whe e a wee is pos ed, i s con en and di e en size and loca ion o clus e s. 1.1 Mo i a ion O e he yea s, Twi e has become one he majo social ne wo ks o sha e ideas and in o ma ion. I con ains coun less da a ega ding people’s in e es s, due o he la ge numbe o use s. This makes Twi e a pe ec se ice o collec da a, p o iding esea che s he necessa y in o ma ion o he de elopmen o da a analysis and knowledge ex ac ion ools. Since he majo i y o in o ma ion sha ed is in ex o ma , Twi e pos s ha e been he sou ce o da a used by ools such as TweeP o iles [17] [18] and TweeP o iles2 [10]. Howe e , i allows sha ing pic u es di ec ly o h ough se ices such as Twi pic3o Ins ag am4. These images can also be used o da a analysis, due o he ac ha in some cases i may be complemen ing he ex o e en eplacing i . The analysis o isual in o ma ion is impo an and enabled he de elopmen o he TweeP o iles’ ex ension Olhó-passa inho [19]. Ou mo i a ion lies wi h he design o an ex ension o TweeP o iles2. One big con ibu ion wi h his p ojec is he de elopmen o a isualiza ion sys em wi h an in e ac i e ep esen a ion o he p o iles as well as he messages hemsel es, connec ed wi h o he pla o ms om SAPO Labs 5 3Mo e in o ma ion a : h ps:// wi pic.com 4Mo e in o ma ion a : h ps://ins ag am.com 5Mo e in o ma ion a : h p://labs.sapo.p 1.2 Objec i es 3 and he e alua ion o his ool in a eal-wo ld applica ion, as well as he in eg a ion wi h a eal- ime da a ex ac ion pla o m named SocialBus6[22]. 1.2 Objec i es This disse a ion aims o e alua e he ool in a eal-wo ld applica ion by de eloping an ex ension o TweeP o iles2 . The goals a e: 1) o comple e he wo k done o TweeP o iles2 ega ding he social dimension; 2) o comple e he in eg a ion o TweeP o iles2 wi h SocialBus; 3) o de elop an in e ac i e isualiza ion ool o displaying p o iles and wee s ound as well as a connec ion o in o ma ion in SAPO7. 1.3 S uc u e This disse a ion is o ganized as ollows: Chap e 2con ains he s a e o he a o he scien i ic hemes ela ed o his p ojec , namely some s eam clus e ing algo i hms and dis ance measu es o each dimension in spa io- empo al Da a Mining, alongside a s a e o he a s udy on spa io- empo al isualiza ion echniques o use in his ool. In Chap e 3we explain he whole a chi ec u e o TweeP o iles3, de ailing he in eg a ion wi h Social Bus, he su ey done o he sys em equi emen s and i s implemen a ion. In Chap e 4we p esen he es ing se up, de ailing some pa ame e s in he clus e ing algo i hm. A ew examples o he esul s ha we e ob ained a e p esen ed and analysed in o de o p o ide some alida ion and o illus a e he ype o knowledge ha can be ex ac ed. I also p esen s he inal su ey done wi h some jou nalis s whe e a usabili y es was pe o med. Finally, in Chap e 5we discuss he esul s ob ained and some o he decisions made h oughou he de elopmen . The mos impo an limi a ions a e discussed and u u e wo k is e iewed. 6Mo e in o ma ion a : h ps:// eac ion. e.up.p /socialbus 7Mo e in o ma ion a : h p://sapo.p 4In oduc ion Chap e 2 S a e o he A This chap e summa izes he esul s o he s udy done o acqui e he equi ed skills and knowl- edge o de elop his wo k. I is ocused p ima ily in isualiza ion echniques bu a b ie analysis o da a mining me hods, is also p esen ed. In Sec ion 2.1, he echnical aspec s ela ed o clus e ing p ocesses and algo i hms, and he simila i y (dis ance) unc ions a e exposed. In Sec ion 2.2 we de ail ele an echniques in isualiza ion sys ems ega ding social media. Finally, an o e iew o Twi e and TweeP o iles is done (Sec ion 2.3). 2.1 Clus e ing Da a mining is he p ocess o explo ing la ge amoun s o da a wi h he goal o inding "in e es ing" pa e ns [23]. Da a Mining is a mul i-disciplina y ield a he con luence o S a is ics, Compu e Science, Machine Lea ning, A i icial In elligence (AI), Da abase Technology, and Pa e n Recog- ni ion, as [24] claims. This p ocess is de ined by se e al asks, depending on he p oblem being add essed, such as [25]: •De ec ion o anomalies (ou lie s / changes / de ia ions) - Iden i y eco ds o unusual da a. Those may be e o s in he da a o in e es ing objec s ha exhibi di e en beha iou om he mos ypical one; •Mining o equen pa e ns, associa ions and co ela ions - Finds pa e ns ha occu equen ly in he da a; •Classi ica ion - Lea ns a model mapping he alues o inpu (independen ) a iables wi h an nominal ou pu (dependen a iable o a ge a ibu e) and applies i o new da a, being mainly used in asks o o ecas ing; •Reg ession - Simila o classi ica ion, excep ha he a ge a iable is nume ic; •Summa y - Compac ep esen a ion o a da a se , which may include desc ip ion and isu- aliza ion h ough a epo ; 5 12 S a e o he A whe e eq( Wi)is he absolu e equency o he e m Wiand Nis he size o he ex .The -id o a e m can be calcula ed by he ollowing o mula: TFIDF =TF( Wi)×IDF( Wi) whe e TF( Wi)is he e m- equency o e m Wiand IDF( Wi)is he in e se documen equency o e m Wi. The in e se documen equency is a s a is ic ha e lec s he impo ance o a e m ela i e o a collec ion o ex documen s. I is gi en by he o mula: IDF( Wi) = logND DF( Wi) whe e ND is he numbe o documen s in he collec ion and DF( Wi)is he numbe o documen s ha he e m Wiappea s in. Le xand ybe wo e m- equency ec o s. The cosine simila i y o hese ec o s gi en by: sim(x,y) = x×y ||x||||y|| wi h ||x|| and ||y|| being he no m o xand y, espec i ely. This measu e calcula es he cosine o he angle be ween he wo ec o s, meaning ha he close he angle is o 90, he mo e unlikely hey ma ch. The alue ob ained will be be ween 0 and 1, 0 meaning ha he documen s ha e no ma ch, and 1 meaning he documen s a e exac ly equal. When he a ibu es o he ec o s e alua ed a e bina y- alued (when i is only conside ed whe he a wo d appea s o no in a ex ) he cosine simila i y unc ion can be in e p e ed in e ms o sha ed a ibu es. A a ia ion o he cosine simila i y in his case is he Tanimo o coe icien : sim(x,y) = x×y x×x+y×y−x×y This o mula gi es us he a io o he numbe o sha ed a ibu es be ween xand y o he o al numbe o a ibu es. 2.2 Visualiza ion An impo an s ep in he Da a Mining p ocess is he in e p e a ion o he esul s. I is o en based on ools o isualize bo h he da a and he knowledge ex ac ed. The main p ope ies ha mus be e i ied by hese ools a e: he displaying o he da a and empo al beha iou ; showing p ope ies o he en i e displayed scene and suppo in e ac ion [36]. Visualiza ion ools in a la ge mul idisci- plina y ini ia i e equi e a p agma ic ye somewha c i ical e iew o he ways isualiza ion can be used o ep esen and o analyse da a [20]. In his p ojec , he e iew o ela ed wo k is ocused on was done based on sys ems ha collec da a om Twi e . A i s , we iden i y some o he di e en dimensions hese sys ems can ake place in. These dimensions, such as he geog aphical coo di- na es (spa ial dimension), he imes amps ( empo al dimension), he connec ions be ween use s (social dimension) and he ex (con en dimension), p o ide he con ex be ween all he da a in he sys em. We also e iew some o he echnology en i onmen s and pla o ms ha a e a ailable. 2.2 Visualiza ion 13 These p esen he playg ound and con ex o he in e ac ion be ween he use , he da a, and he isualiza ion p oduc . The goal is o iden i y he possible in e ac ion pa adigms o be suppo ed by he isualiza ion ools ha de eloped in his p ojec . Finally, we look a some echniques ha can be used o isualize geospa ial da a. Ins ead o adi ional isualiza ion echniques like g aphs and cha s, we analyze a se o con empo a y geo- isualiza ion echniques made o imp o e he expe ience. Besides di e en echniques, di e en ypes o in o ma ion a e displayed, enabling he design o isualiza ion ools based on clus e ing and/o geo e e enced da a. 2.2.1 Clus e ing Visualiza ion Fo clus e ing isualiza ion, he mos common ep esen a ion a e g aphs. The objec s in each clus- e a e p esen ed and he goal o assigning simila objec s he sho es dis ance be ween clus e s is main ained. A sys em de eloped by [32] can be seen in Figu e 2.3. Figu e 2.3: Clus e ing Visualiza ion Ano he clus e ing isualiza ion me hod o a la ge amoun o da a in ol es assigning di e en colo s and objec s. Fo objec s in di e en clus e s, o e lapping ellipses o e he mos ep esen- a i e objec s a e displayed o ep esen simila objec s [17] [18]. This app oach was applied o s udy geog aphical lexical a ia ion ( igu e 2.4a [33]) and o classi y e en s ( igu e 2.4b [31]). (a) Clus e ing Visualiza ion. (b) Clus e ed dis i- bu ion a ound i [31]. Figu e 2.4: Clus e ing Visualiza ion Examples. 14 S a e o he A Mo e ecen ly di e en clus e ing isualiza ion me hods we e designed based on a di e en echnique. In [3] clus e s a e displayed in a mo e inno a i e and in e ac i e way, enabling he use o easily iden i y con en in each clus e . An example o his sys em can be seen in Figu e 2.5. Figu e 2.5: Clus e ing Visualiza ion o he wee s o he sea ch e m " echnology" on May 16, 2013 [3]. 2.2.2 Visualiza ion Me hods Since he Twi e ’s API p o ides di e en ypes o in o ma ion such as wee ID, use ID, use name, imes amps, la i ude, longi ude and ex , mul iple isualiza ion echniques can be used o isualize i . The ollowing lis b ie ly discusses some o hese echniques. •Cho ople h Maps: A cho ople h map ep esen s agg ega ed measu es o p e-de ined ge- og aphical egions (W igh , 1938). Cho ople h mapping is pa icula ly use ul in p o iding compa a i e summa ies o e speci ied geog aphies [20]; •Hea Map: Also known as densi y su aces, a hea map is a isualiza ion echnique o ep- esen he densi y o spa ial da a using Ke nel Densi y Es ima ion me hods [20]. I enables use s o easily iden i y high densi y a eas wi hou losing he gene al spa ial con ex . I is ypically used o analyse geospa ial da a, whe e a sense o co ela ion be ween geog aphical ea u es and o he measu emen s is equi ed. Hea maps a e also use ul in mapping empo al u ban phenomena such as people and a ic lows [37]; •Flow Map: Flow maps display mo emen s o objec s o subjec s om one place o ano he by means o lines o a ows [20]. The da a used needs di e en ini ial and inal geog aphical loca ions, such as, o example, mig a ion pa e ns be ween egions. Al hough hey a e usually s a ic, low maps can be made dynamic by using a ime sequence anima ion [20]; •Social Map: A social map is he ca og aphic, wo-dimensional ep esen a ion o social in e ac ions. In sys ems such as [3] i is used as a ep esen a ion o clus e s con aining simila in o ma ion (in his case simila wo ds om wee s); 2.2 Visualiza ion 15 •B ushing: Also known as mul iple-linked iews, i is a me hod o dynamic que ying by di ec manipula ion o isual and da a displays wi h he esul s being upda ed based on manipula ion is commonly e e ed o as "b ushing" [20]. I is usually used o explo a o y da a analysis [38]. Gahegan [36] claims ha he main isualiza ion echniques a e: map-based, cha -based, p o- jec ion, space- illing o pixel based, iconog aphical o composi ional and hie a chical o ne wo k. This is no in ended as an exhaus i e e iew o all he ele an echniques. 2.2.2.1 Geo e e enced Da a Visualiza ion Geo e e enced da a in ol es displaying he in o ma ion on a geog aphic ep esen a ion, usually a map (such as Google Maps1, Google Ea h2o NASA Wo ldWind3). One o he sys em c ea ed by Google ha enables a 3D isualiza ion o he Ea h is Google Ea h. Wi h an easy in e ace, i allows an in ui i e ep esen a ion o geo e e enced da a.. Google Maps is a e y popula 2D map isualiza ion ool. A la ge numbe o applica ions we e de eloped using hei API because o i s simplici y and isual appeal, including some o he ools desc ibed ea lie [6,39,4]. Figu e 2.6: Map wi h e en de ec ion on Twi e [4] E en hough only isualiza ion ools om Google ha e been men ioned, he e a e many o he s in his ma ke ha also p o ide a map API [17] [18]. Silicon G aphics In e na ional, in pa ne ship wi h he Uni e si y o Illinois designed a eal- ime isualiza ion pla o m o sen imen mining on Twi e [5]. The ool adop ed a hea map ep esen- a ion, in which each colo ep esen ed a di e en alue o he majo i y o posi i e o nega i e commen s [17] [18]. 1h ps://de elope s.google.com/maps/?hl=p -p 2h ps://de elope s.google.com/ea h/?hl=p -p 3h p://wo ldwind.a c.nasa.go / 16 S a e o he A Figu e 2.7: Real- ime hea maps o posi i e and nega i e sen imen s exp essed ia Twi e [5] 2.2.3 Twi e Da a Visualiza ion An analysis o se e al ools o isualiza ion o Twi e da a was made. In able 2.2 some o hose ools, in o de o unde s and how cu en sys ems en ision using geo isual ools can ake ad an- age o social media da a sou ces (namely Twi e ). Twee Pos SensePlace2 Compac Map Ea hquake Twee D ops A So wa e Sys em o Da a Mining wi h Twi e E en Rada Dimensions Spa ial, Tem- po al, Con en (Hash ags) Spa ial, Tem- po al, Con- en Con en Spa ial, Tem- po al, Con- en Con en Spa ial, Con en Spa ial, Tem- po al, Con- en En i onmen Digi al Globes Digi al Globes N/A Digi al Globes N/A Digi al Globes Digi al Globes Pla o m Google maps P op ie a y N/A N/A N/A Google maps N/A Visualiza ion Techniques Hea Map/ G aph & Cha s Hea Map Social Map Hea Map/ G aph & Cha s Rain d ops B ushing/Hea Map/ G aph & Cha s B ushing Type o i- sualiza ion 2D A ea 2D A ea N/A 2D A ea 2D Poin s 2D A ea 2D A ea Type o in o ma ion displayed Densi y o wee s Posi ion and Con en o wee s Size o Clus e s/ Con en o wee s Posi ion o wee s Con en o wee s Posi ion and Con en o wee s/ Densi y o wee s Posi ion o wee s S eam han- dling Online Online Online O line O line O line Online Clus e ing algo i hm N/A N/A LDA (La en Di ichle Alloca- ion) N/A N/A N/A DBSCAN Table 2.2: Di e ences be ween Visualiza ion Sys ems Twee Pos Twee Pos [6] is a web se ice ha is in ended o acili a e he analy ical s udy o geog aphic endencies in Twi e da a eeds. In o de o imp o e he use ’s expe ience wi h he ool, Twee Pos elies on isual da a s uc u es like hea maps and cha s o ep esen he geo-spa ial sou ces o wee s. Figu e 2.8 displays an hea map example o he Twee Pos ool. 2.2 Visualiza ion 17 Figu e 2.8: Hea map-based as well as quan i a i e compa ison o game console popula i y om [6]. SensePlace2 SensePlace2 [7] p esen s a geo isual analy ics app oach o suppo si ua ional awa eness (SA) o c isis e en s using Twi e . I ocuses on le e aging explici and implici geog aphical in o ma- ion o wee s and on p o iding isual in e ace me hods o enable unde s anding o place, ime, and heme componen s o e ol ing si ua ions [7]. I is a use -cen e ed app oach, using scena io-based designs ha include o mal scena ios o guide and alida e implemen a ion as well as a sys ema ic claims analysis o jus i y design choices and p o ide a amewo k o u u e es ing. I is composed by a s uc u ed su ey o p ac i ione s and he end p oduc o Phase-I de elop- men is demons a ed h ough implemen a ion o a map-based, web applica ion ini ially ocused on wee s bu ex ensible o o he media [7]. Figu e 2.9: Visualiza ion o que y es ic ed o wee s wi h geo-loca ion om [7] Twee D ops Twee D ops [8] is a compu e -based isualiza ion ool designed o people who ha e no paid 18 S a e o he A a en ion o sus ainabili y in hei li e be o e. I opens up an oppo uni y o hem o lea n abou ene gy conse a ion. I has wo main isual componen s, as shown in Figu e 2.10. One is he backg ound ain d ops, which ep esen he accumula ion o ene gy ela ed wee s collec ed om Twi e ; he o he is clickable o eg ound wee s wi h de ailed con en . Figu e 2.10: Fo eg ound clickable wee s a e displayed as la ge blue sen ences in on o he ain d ops [8]. Compac Map Compac Map [3] is an online isual in e ace ha packs ex clus e s e icien ly. I achie es spa io- empo ally cohe en layou s by dynamically ma ching clus e s ac oss ime, and emo ing clus e o e laps acco ding o spa ial p oximi y and cons ain s. Compac Map enables: •A dynamic isualiza ion echnique ha displays clus e s in ex s eams as s able, space- e icien layou s; •A eal- ime isual sea ch engine ha suppo s a bi a y keywo d sea ch combined wi h se- man ic analysis o opics; •An enhanced eal- ime isual analysis sys em ha enables use s o explo e and compa e discussions and opics. A p e iew o Compac Map can be seen in Figu e 2.5. Ea hquake Ea hquake [9] is a isualiza ion ool ha uses Twi e pos s ega ding he ea hquake which oc- cu ed on he Eas Coas o he Uni ed S a es (US) on Augus 23, 2011. I ga he s in o ma ion based on hash ags and displays he loca ions o di e en wee s in di e en ime pe iods. I dis- plays in o ma ion as hea maps as well as g aphs. An example o his ool can be seen in Figu e 2.11. 2.2 Visualiza ion 19 Figu e 2.11: Ea hquake isualiza ion ool o 40 wee s [9]. A So wa e Sys em o Da a Mining wi h Twi e A So wa e Sys em o Da a Mining wi h Twi e [39] is a isualiza ion sys em designed o display Twi e in o ma ion based on di e en que ies. I ga he s da a pos ed om a ious geog aphic e- gions and hen e ie es ele an wee s om ha da a using ei he spa ial o ex ual que ies. I enables he use o inspec e ie ed wee s using a map based in e ace and g aph/cha s. Figu es 2.13a and 2.13b gi e examples o analysis ha can be done wi h his ool. (a) [39] map isualiza ion (b) Dis ibu ion o wee s con- aining e e ences o Smoking [39]. Figu e 2.12: A So wa e Sys em o Da a Mining wi h Twi e . E en Rada E en Rada [31] is a no el local e en de ec ion me hod ha analysis a se en day his o ic wee da a in o de o imp o e p ecision. I uses maps isualiza ion echniques in o de o display he posi ion o ele an wee s ega ding ce ain e en s. 20 S a e o he A (a) A sample clien using E en Rada [31]: a map showing local e en s. (b) A sample clien using E en Rada [31]: de ails o a selec ed local e en . Figu e 2.13: E en Rada . 2.3 The TweeP o iles P ojec TweeP o iles is a ool o mul i-dimensional clus e ing and isualiza ion o Twi e da a. I ana- lyzes geog aphical coo dina es (spa ial dimension), imes amp ( empo al dimension), use s (social dimension) and ex (con en dimension). This sec ion gi es a b ie o e iew o Twi e as well as a simple desc ip ion o TweeP o iles [17,18] and i s a ian s, Olhó-Passa inho [19] and o TweeP o iles2 [10]. 2.3.1 Twi e Twi e 4is a social ne wo king and mic oblogging se ice. I allows he use s o b oadcas sho ex messages wi h 140 cha ac e s in leng h, known as " wee s", o iends o " ollowe s". Figu e 2.14 shows an example o a wee . Figu e 2.14: Example o a wee . Twi e is abou disco e ing in e es ing people online and ollowing hei bu s messages o as long as hey a e in e es ing. In Twi e , a social connec ion is made when a use ollows o is ollowed by ano he one. A wee is composed by di e en i ems and each may se e di e en pu poses. These i ems can be o social in e ac ion o o comple e he in o ma ion ela ed o he message in ques ion. 4h ps:// wi e .com 2.3 The TweeP o iles P ojec 21 •Re wee (RT) Sha e ano he use ’s wee ; •Men ion (@ + use name) Iden i y a use in a wee ; •Reply (@ + use name) Answe o a p e ious use wee ; •Hash ag (# + opic name) Associa ion o a keywo d o a wee ; •Localiza ion Use ’s geo-coo dina es when sending he a wee . 2.3.1.1 Twi e API Twi e ’s in o ma ion can be accessed by he use o wo APIs: he REST API [40] and he S eam- ing API [41]. The i s is eques -based and equi es an "oAu h" au hen ica ion while he la e uses e en s o p o ide in o ma ion and equi es ei he "oAu h" o HTTP basic au hen ica ion. The REST API allows access o in o ma ion abou he use , he imeline, iends and ollowe s, di ec messages geoloca ion, ends, and mo e. Howe e , his API limi s he numbe o eques s allowed pe use o ei he 15 o 180 wee s in a 15 minu e window, depending on he me hod ha is u i- lized. I is also limi ed by he a ailabili y o Twi e da a and applica ions a e limi s. The p ocess o connec ion and da a e ie al o he REST API can be seen in Figu e 2.15. Figu e 2.15: In e ac ion p ocess o Twi e ’s REST API The S eaming API uses eal- ime da a, limi ing he amoun o da a a ailable acco ding o he session’s begin imes amp. This da a is accessed h ough h ee ypes o s eams: Public, Use and Si e. The i s in ol es he public da a a ailable on his social ne wo k. In he o he s, il e s a e applied in o de o only isualize wee s om ce ain use s. Figu e 2.16 shows he s eaming API p ocess o connec ion and da a e ie al. 2.3.2 Máquina do Tempo Máquina do Tempo ( ime machine) is an in e ac i e ool ha allows he na iga ion and explo a ion o news, om he las 25 yea s, o he Po uguese news agency (LUSA) ile as well as he main websi es o Po uguese news. Wi h his sys em use s can ake a ip back in ime, e isi ing he mos memo able e en s and pe sonali ies o Po uguese and In e na ional his o y in ecen yea s, 28 S a e o he A Chap e 3 TweeP o iles3 In his Chap e i is explained he en i e a chi ec u e o he de eloped ool, TweeP o iles3. In he ollowing sec ions we p esen he esul s o he p ac i ione su ey, p o ide an o e iew o he applica ion and i s a chi ec u e ocusing on he in eg a ion be ween SocialBus and TweeP o iles2 and he di e en pla o ms om Sapo. 3.1 In oduc ion As men ioned be o e, TweeP o iles2 was de eloped as an ex ension o TweeP o iles, allowing clus e ing in eal ime. The clus e ing is done o e h ee dimensions: spa ial, empo al and con- en dimension. In o de o ob ain da a om Twi e in eal ime, we in eg a ed TweeP o iles2 in Social Bus. So- cialBus [22] is a pla o m de eloped in Sapo Labs and i allows use s o ga he da a om Twi e and s o e he desi ed in o ma ion. This ool will be explained in mo e de ail in Sec ion 3.2. In eg a ing hese wo ools is a big imp o emen o TweeP o iles2, due o he ac ha we may es ic SocialBus o sa e only wee s wi h he equi ed in o ma ion o he ask. Also, no only i allows TweeP o iles2 o be pe o med di ec ly om he Twi e s eam in eal ime, bu also allows he in o ma ion o be sa ed, in o de o apply clus e ing o he same da a. This is an impo an ea u e since his p ojec is expec ed o be con inued in he u u e wi h he objec i e o imp o ing he algo i hms used. The key inno a ion p esen ed he e is he in eg a ion o linked isual-compu a ional me hods and a place- ime-con en concep ual amewo k in a wo king p o o ype g ounded in bo h heo y and p ac ice in o med by a s uc u ed su ey o p o essionals. 3.2 Sys em A chi ec u e Since one goal o his p ojec was o in eg a e TweeP o iles2 in SocialBus, in o de o de ine he a chi ec u e o his sys em i was necessa y o unde s and he a chi ec u e behind hese wo sys- ems. Sec ion ?? in oduced he a chi ec u e o TweeP o iles2 and i is ep esen ed in igu e 2.19. 29 30 TweeP o iles3 SocialBus is a esea ch pla o m o ex ac ing, s o ing and analysing he Twi osphe e o R& D and jou nalis ic pu poses. I s a chi ec u e is p esen ed in igu e 3.1. SocialBus collec s wee s in eal- ime using Twi e ’s S eaming API. These wee s a e sen o a message b oke (i.e., da a o ma ansla o p og am) and p ocessed o e wo componen s: s eam p ocessing and p e- p ocessing. The esul ing da a is s o ed MongoDB. Figu e 3.1 ep esen s he comple e sys em bu only he mos impo an s eps we e men ioned. Figu e 3.1: SocialBus high-le el a chi ecu e. TweeP o iles3 was de ined by combining hese wo sys ems. We use SocialBus’ ex ac ion and p e-p ocessing me hods o ga he da a om Twi e and included TweeP o iles2 in he sys em so ha when he da a is s o ed, i au oma ically pe o ms clus e ing. These esul s a e hen s o ed and mapped. The use case diag am in igu e 3.2 illus a es how use s a e expec ed o in e ac wi h he sys em. TweeP o iles3 mus : •Allow use s o collec Twi e da a pos ed om a ious geog aphic egions; •Allow use s o e ie e ele an wee s and clus e s om he collec ed da a using spa ial que ies; •Allow use s o e ie e ele an wee s and clus e s om he collec ed da a using ex ual que ies; 3.2 Sys em A chi ec u e 31 Figu e 3.2: TweeP o iles3 use case diag am. •Allow use s o e ie e ele an wee s and clus e s om he collec ed da a using empo al que ies; •Allow use s o inspec e ie ed wee s and clus e s using a map-based in e ace. These equi emen s a e ul illed by se e al so wa e modules ha a e de ailed he e. Da a Collec ion Module The design o he da a collec ion so wa e module sepa a es web-in e ace logic om he back- g ound p ocesses o da a collec ion and s o age. This is comple ely done by SocialBus who uses he open-sou ce Twi e 4J lib a y o connec ion o he S eaming API and s o es ecei ed wee s in MongoDB. The web in e ace le s use s moni o hei p og ess. The a chi ec u e o his module was p esen ed p e iously in igu e 3.1. 32 TweeP o iles3 Da a S o age Module The collec ed da a is ini ially s o ed in MongoDB. The geo-loca ed (la i ude/longi ude coo - dina es) and empo al in o ma ion embedded in wee s mus be ep esen ed in he co ec o ma be o e sea ch que ies and dis ance measu es can be applied on he da a. The da a collec ed om he S eaming su e s a p e-p ocessing done by SocialBus in o de o con- e da e in o ma ion o a usable o ma con aining "yea -mon h-day : hou -minu e-second". Da a Re ie al and Clus e ing Module Da a Re ie al and Clus e ing a e bo h in eg a ed in he same module. Design and implemen a ion o he Da a Re ie al module a e s aigh o wa d a e he con e sion o Twi e da a. MongoDB p o ides easy me hods o e ie e s o ed da a. Since no all he col- lec ed con ains he equi ed in o ma ion o he clus e ing algo i hms, in o de o ex end he spa ial and empo al e ie al unc ionali y o e ed o end use s, some o he p e-p ocessing was equi ed. This is will be de ailed in Sec ion 3.4. As he da a is e ie ed om MongoDB i is passed o he clus e ing algo i hms, DenS eam and DBSCAN. DenS eam is esponsible o he c ea ion o mic o-clus e ha a e hen used as inpu o DBSCNAN in o de o c ea e he mac o-clus e s. Bo h o hese algo i hms we e implemen ed in Ja a. Sea ch Module Full- ex sea ch is inco po a ed in ou sys ems o assis end use s wi h ex ual analysis, speed- up que ies on a la ge da a-se and o p oduce a b oade se o sea ch esul s o each keywo d speci ied by he use . The sys em p e o ms a ma ch o he desi ed wo d o bo h wee s and se o wo ds in all clus e s. The sys em does no p esen an app oach o deal wi h slang wo ds and abb e ia ions commonly ound in Twi e con en . Since Twi e en o ces a s ic 140 cha ac e limi o each wee , ab- b e ia ions and sho o ms o many wo ds and ph ases a e widely used o p ese e space. Fo example, he slang abb e ia ion ’IMHO’ is used o sho ening he ph ase ’in my hones opinion’. As he majo i y o hese abb e ia ions a e no in dic iona ies, hey p esen a subs an ial challenge owa ds ai h ul analysis o Twi e con en using na u al language p ocessing and da a mining echniques. The sys em also inco po a es a ull spa ial sea ch enabling he use s he de ine a speci ic egion o analyse, as well as a empo al sea ch so ha he use can de ine a ce ain weekday o display all wee s and clus e s and a imeline o go back in ime 7 days. Da a Mapping Module 3.2 Sys em A chi ec u e 33 The inal module o TweeP o iles3 is used o da a isualiza ion. I is esponsible o ou - pu ing he esul s o he clus e ing algo i hms and o spa ial, empo al and ex ual que ies o end use s. The Google Maps and Lea le 1Ja asc ip API a e used o display e ie ed wee s and esul ing clus e s on a map. Figu e 3.3: Modules om TweeP o iles3 3.2.1 Unde s anding Use Needs As an inpu o ou ool design and de elopmen p ocess, we comple ed a needs and in e es s udy wi h wo p ima y goals: (1) o de elop a clea unde s anding o how media p o essionals use social media ools and (2) o d aw ou wha jou nalis en ision doing wi h social media geo isual ana- ly ic ools in he u u e. To add ess hese objec i es, a 32 ques ion su ey deli e ed online using Google Docs was c ea ed ( his can be seen in Appendix A). To ec ui pa icipan s we e con ac ed people in he media ield, namely people om FEUP, Jo nalismoPo oNe (JPN) and Publico P3. These in e iews yielded 10 esponden s o ou su ey. This su ey can be di ided in h ee subg oups. One whe e we ga he ed in o ma ion abou he backg ound o each pa icipan , ano he whe e we asked abou he use and knowledge o social media and he las whe e we asked pa icipan s o compa e di e en applica ions simila o wha we had planned. The i s ques ion asked abou wo k expe ience in media. All esponden s we e people wi h ex- pe ise and he majo i y had o e 2 yea o expe ience. The ollowing Sec ion e eals wha we lea ned om he su ey on ou goals. The esul s a e p esen ed wi h pe cen ages o ques ions ha allowed mul iple op ions o be selec ed, and by a e age subjec i e a ing o ques ions using a 5-poin Like ag ee/disag ee scale. 1h p://lea le js.com 34 TweeP o iles3 3.2.2 Use o Social Media and Geo isual Tools To begin unde s a ing how social media is used by p o essionals, we asked pa icipan s o desc ibe hei use o social media ools in pe sonal and p o essional en i onmen s. Facebook (100%), Twi - e (100%) and Linkedin (80%) a e mos common o pe sonal usage. In e ms o p o essional usage, Facebook (100%) and Linkedin (60%) a e mos common. When i comes o jou nalism, Facebook (100%) and ex messages (80%) a e p ima ily used by he pa icipan s o communica e wi h o he p o essionals, Facebook (100%) and Twi e (60%) o announce and ga he in o ma ion om he public, and Facebook (100%) and ex messages (40%) o communica e wi h he public. I was also asked he esponden s wha aspec s o ools like Twi e hey alued he mos , and less, as a suppo pla o m o social media. The immediacy and in e ac i i y we e he aspec s ha pa icipan s desc ibed as he mos impo an and he possibili y ha , in he chaos o such pa allel in o ma ion, ollowed by use s, he jou nalis ic message is los , was he nega i e one. Wi h ha in mind, we ied o unde s and how use ul hese ools a e, in ega d o aspec such as ins an anei y, in e ac i i y, pe pe ui y, mul imedia and hype ex uali y. All pa icipan s ag eed ha Twi e is use ul as a jou nalism ool o e ie e and sha e in o ma ion due o ins an anei y. In he o he majo pa o ou su ey, we ied o unde s and how he pa icipan s en isioned using geo isual ools ha ake ad an age o social media da a sou ces, namely Twi e . To begin, we asked pa icipan s o iden i y ypes o maps hey would expec o see in new social media ools. They indica ed ha he loca ions men ioned in o ele an o he con ibu ed in o ma ion (100%) a e mo e use ul han he loca ion o he pe son gene a ing he in o ma ion. Addi ionally, pa ici- pan s esponded ha hey would ind maps ha show bo h ypes o geog aphic in o ma ion a once o be be e han jus one. I was hen asked o he pa icipan s o iden i y ce ain ools and ea u es expec ed in an in e ac- i e web-based applica ion o inco po a e social ne wo ks as a da a sou ce o media. Pho os/ ideo collec ions (100%), ables (60%), maps (60%), keywo d clouds (60%) we e among he mos popu- la choices. A e ha we asked he pa icipan s o iden i y ypes o in o ma ion hey would expec o be able o analyze in he same applica ion. Sequence o e en s (100%), people (100%), o gani- za ions (100%) and inciden s (60%) we e he mos expec ed in o ma ion ypes. In a inal se ies o ques ions, we asked hem o e iew a g aphical mockup o he new TweeP o iles in e ace. Fi s , we asked pa icipan s o desc ibe how hey migh use such an applica ion. Some o he answe we ob ained we e: •I’d choose he loca ion, in o de o see wee s om he e, and mos equen wo ds and ela ed news. •I’d use he in o ma ion ha was mo e p e alen and equen , econciling he map, he in o ma ion o he clus e s and wee s. 3.3 Visualiza ion Sys em 35 A e seeing he mockup, we demons a ed wo o he applica ions de eloped in Sapo Labs, he i s TweeP o iles 2and Re wee Pa e n3, and asked he pa icipan s o compa ed hem wi h he in e ace shown. Nex , we asked he esponden s wha ypes o in o ma ion o esul s hey would expec o be able o sha e i hey we e o use an applica ion like TweeP o iles3. This e ealed ha p e- o ma ed ex epo s (60%), p in able maps (40%), a link ha would launch he applica ion wi h p eloaded da a (40%) and s a ic sc een cap u es (40%) we e p e e ed. The esul s o ou su ey sugges ha hose in social media, namely jou nalism, a e ac i ely using social media ools, ha hey expec o be able o explo e mul iple kinds o geog aphic in o ma ion and an icipa e o make use o ha in o ma ion. Ou esul s p o ide a ack o cu en and u- u e designs o TweeP o iles ha include unc ionali ies o mapping ools, media in eg a ion and analy ical epo ing capabili ies. 3.3 Visualiza ion Sys em This Sec ion in oduces ou applica ion design p ocess and ini ial sys em implemen a ion and es ing. Fi s , we ou line a base scena io, based on equi emen s and echnologies, ini ial TweeP- o iles3 unc ionali y and p esen a pla o m analysis ocused on key design choices. 3.3.1 Base Scena io Ou ool and in e ace design app oach ollows he esul s om ou p ac i ione su ey and unde - s anding o use s’ needs. Figu e 3.4: Response abou ools/ ea u es expec ed in TweeP o iles3 Based on ha knowledge, as well as he inqui y done o he s a e o he a , we de ined a se o echnologies o inco po a e he majo i y o expec ed ea u es and in o ma ions in a web applica ion o suppo jou nalism. Displaying a map wi h wee s’ and clus e s’ in o ma ion is a mus , o an applica ion o his ype. 2h p:// eac ion. e.up.p / weep o iles/ weep o iles.h ml 3h p:// od igues37.gi hub.io/Re wee Pa e n 36 TweeP o iles3 Figu e 3.5: Response abou in o ma ions expec ed in TweeP o iles3 We decide o use Google Maps and Lea le , due o he ac ha bo h a e easy and accessible o use and con ain all ea u es equi ed o he asks planned. Since wo dclouds and ime g aphs we e among he expec ed ea u es, we used wo plug-ins om d3 4 o display ha in o ma ion. 3.3.2 TweeP o iles3 Func ionali y TweeP o iles3 exposes i s unc ionali y o end use s h ough a php web applica ion amewo k, designa ed Codeigni e 5. I also p o ides a simple web based da a collec ion and e ie al in e - ace, which is shown in igu e 3.6. Figu e 3.6: Sc eensho o TweeP o iles3’s web in e ace. The sys em in e ace includes 5 co e componen s: 1) Twee and clus e map; 2) wee , news and en i ies lis ; 3) display/dimension con ols; 4) wo dcloud; 5) ime g aphics. Each componen is desc ibed in mo e de ail below. 4h p://d3js.o g 5h p://www.codeigni e .com 3.4 Sys em Implemen a ion 37 Twee and Clus e map: Following he logic ou lined p e iously, he map suppo s simul ane- ous wee s and clus e ing dis ibu ion o e iew. In one pla o m, a hea map p o ides he o e iew o wee s and he clus e s a e displayed using simple ma ke s. In a second pla o m, wee s’ po- si ion is displayed using ma ke s and clus e s wi h ci cles ( igu e 3.6). He e we can ge de ailed in o ma ion such as he numbe o wo ds in he clus e , he c ea ion ime and posi ion. Twee , News and En i ies lis : The wee lis depic s he 1000 newes wee s o any que y. The en i y lis displays any pe sonali y ound in he p ocessed wee s and he news lis display he 20 mo e ecen news o hose en i ies om Sapo. Display/Dimension con ols: The Display con ols suppo que y il e ing wi h egion selec- ion, ime ange slide s (ei he choosing weekday o going back in ime) and a minimalis ic in- e ace o speci y e ms o ex -based que ies. These con ols also include a il e o speci y he desi ed dimensions. Figu e 3.7: Sc eensho o TweeP o iles3’s il e s in e ace. Wo dcloud: The wo dcloud displays he mos equen wo ds in all clus e s. The size o he wo ds is equi alen o i s equency in wee s. The mo e i appea s, he bigge he wo d’s size is. Time g aphics: These g aphics display he hou and weekday o he c ea ion o all clus e s. The size o he clus e co esponds o he numbe o poin s i con ains. All widge s a e ela ed. When selec ing an elemen in he map, all o he widge s (wo dcloud and ime g aphs) change in ela ion o ha . Besides ha , i a mac o-clus e was selec ed, only mic o-clus e s ela ed o ha mac o will be shown in he map. This can be seen in Sec ion 3.4.3. 3.4 Sys em Implemen a ion In he ollowing Sec ion we p esen he implemen a ion o TweeP o iles3. 3.4.1 SocialBus mee s TweeP o iles2 As men ioned p e iously, one goal o his wo k was o ully in eg a e TweeP o iles2 in SocialBus. Bo h o hese pla o ms we e de eloped en i ely using he Ja a language o all he algo i hm and p ocesses wi h MySQL and MongoDB da abase o s o e in o ma ion. These wo da abases we e chosen due o he ac ha MongoDB p o ides easy me hods o access and p ocess he da a s o ed, so i made easie o sa e and analyse wee s ha came om he s eam and MySQL has simple 44 Resul s 4.1 Clus e ing To access he algo i hm, we pe o med a se ies o es ing ounds wi h di e en componen combi- na ion ( o DBSCAN). Table 4.1 shows all he es s done o da a ga he ed. Combina ion Spa ial Dimension Tempo al Di- mension Con en Dimension 1100% 100% 100% 2100% 0% 0% 30% 100% 0% 40% 0% 100% 5100% 100% 0% 60% 100% 100% 7100% 0% 100% Table 4.1: Se o es s pe o med. Since he dis ance unc ion weigh ing was no pe o med co ec ly in TweeP o iles2, we de- cided no o include ha op ion, u ning all dis ance unc ions in o a bina y a iable. Besides ha , Hyb idDenS eam has di e en pa ame e s ha can be al e ed o p oduce di e en esul s o he clus e ing. Name Abb e Desc ip ion Min Value Max Value De aul Value Epsilon eps De ines he mini- mum adius o a HMC 0 1 0.1 MinPoin s mp De ines he min- imum numbe o poin s in he ε- neighbou hood o c ea e a HMC (also used as µpa ame e ) 1∞2 Ini Poin s ip Numbe o poin s o ini ializa ion 50 ∞1000 µ µ Used in he PM- C/OMC es ic ion 1∞1 Be a βUsed in he PM- C/OMC es ic ion 0 1 0.2 Lambda λUsed in he ime de- cay unc ion; a ec s he decay a e o he s eam 0 1 0.25 P ocessing Speed s De ines he numbe o ins ances ( wee s) pe ime uni 1∞100 Table 4.2: Hyb idDenS eam pa ame e s. DBSCAN also has wo inpu pa ame e s, epsilon and MinPoin s. These ha e he same mean- ing as he one men ioned abo e, bu a e used o he mac o-clus e ing p ocess. In o de o ob ain 4.1 Clus e ing 45 (a) Time poin 1 (b) Time poin 2 Figu e 4.3: Spa ial Clus e ing esul s. he esul s shown below, as he main goal o his p ojec was o ex end TweeP o iles2 and no imp o e he obus ness o he clus e ing algo i hm, all hese pa ame e s we e se o de aul alue, besides Ini Poin s ha was se o minimum. DenS eam and DBSCAN’s pa ame e s (eps and Min- Poin s) we e se o 0.4 and 1 ( o DenS eam) and 0.4 and 2 ( o DBSCAN). All he esul s demons a ed below we e ob ained wi h hese alues. Spa ial Dimension The esul s o his es ing ound we e ob ained wi h he second combina ion o able 4.1, mean- ing only he spa ial dis ances we e aken in o accoun . Figu e 4.3 shows he clus e s ob ained o his dimension. In his ep esen a ion, each blue ci cle ep esen s a mic o-clus e and all g een ci cles, he mac o-clus e s. All mic o and mac o clus e s’ adius a e calcula ed based on i s a- dius and he numbe o poin s. Each ime poin is a momen in he simula ion whe e we s o ed he clus e ing esul s. I is use ul o see he e olu ion o he clus e s and in his case, since only spa ial in o ma ion is being p ocessed, how he mac o-clus e s appea in a la ge concen a ion o mic o-clus e s. The esul ing clus e s can be deemed adequa e, due o he ac ha wi h hese languages being p ocessed, mos o he wee s will be coming om Eu ope and Sou h Ame ica. Tempo al Dimension When i comes o empo al dimension, some p oblems we e aced ega ding he clus e ing. Since we used he de aul pa ame e s o bo h algo i hms and he mic o-clus e s a e c ea ed aking all dimensions in o accoun , DBSCAN c ea es mul iple mac o-clus e wi h e y simila imes- amps. E en hough we aced his p oblem, i can clea ly be seen in igu e 4.4 ha he sys em c ea es one clus e including he majo i y o poin s, when he da ase is om a single day only. Figu e 4.5 shows ha big di e ence be ween he esul ing clus e s. 46 Resul s Figu e 4.4: Tempo al Clus e ing esul s. A big di e ence can be seen when including he spa ial dimension o his combina ion. Figu e 4.6 p esen s he esul s o his clus e ing. Con en Dimension Fo his ound o es s, i was only aken in o accoun he con en o he wee (combina ion 4 om able 4.1). The esul s a e no as sa is ac o y as we would wan , due o he ac ha compa ed ex s a e small, he compu ed simila i y is also small. E en hough he ε alues a e high, he numbe o esul ing clus e s o his dimension is small. Howe e , i i inc eased e en mo e he quali y o he esul s may dec ease, because he algo i hm will be me ging ex ins ances ha ha e li le simila i y. This was a p oblem also aced in TweeP- (a) Time poin 1 (b) Time poin 2 Figu e 4.5: Tempo al Clus e ing esul s. 4.1 Clus e ing 47 Figu e 4.6: Tempo al and Spa ial Clus e ing esul s. o iles2. Figu es 4.7,4.8 and 4.9 gi e he esul s o a es o he con en dimension. I p esen s he e olu ion o a clus e o e di e en pe iods o ime. I shows, o each ime poin , he clus e and he mos equen wo ds o med a ha momen . The size o he wo ds is p opo ional o hei equency ela i e o o he s. I can be seen ha , e en hough hey di e in size, he mos common wo ds emain consis en . Like he p e ious dimension, a di e ence can be seen when including spa ial dis ances o he clus e ing. Figu e 4.10 shows how he clus e changes when only using he con en dimension. Figu es 4.6 and 4.10 can also be compa ed o 4.3 whe e a big di e ence can also be seen. E en hough he di e ences a e ha d o e alua e, in some cases i can be seen how he ou come e lec s when including o he dimensions o he clus e ing. All esul s p esen ed demons a e some in e es ing pa e ns. The esul s ob ained we e sa is ac o y in he sense ha i made possible o unde s and be e how he algo i hm eac s o SocialBus’ da a ex ac ion. I p o ides a ully in eg a ed eal- ime clus e ing sys em wi h a s a e o he a isualiza ion pla o m. Wi h he clus e ing being done in eal- ime bu no being op imized, he esul s ob ained he e a e no as good as he esul s in he la es publica ion o TweeP o iles [18]. Figu es 4.11 and 4.12 p esen wo examples ob ained. 48 Resul s Figu e 4.7: Con en Clus e ing esul s. Figu e 4.8: Con en Clus e ing esul s. 4.1 Clus e ing 49 Figu e 4.9: Con en Clus e ing esul s. Figu e 4.10: Con en and Spa ial Clus e ing esul s. 50 Resul s Figu e 4.11: TweeP o iles spa ial clus e ing esul s. Figu e 4.12: TweeP o iles empo al clus e ing esul s. When compa ing hese example o he spa ial ( igu e 4.3) and empo al ( igu e 4.5) clus e ing esul s om TweeP o iles3, i is clea ha TweeP o iles p o ides he bes pe o mance. Howe e , i does no enable a eal- ime da a analysis, which is a big ad an age o ou sys em. 4.2 Usabili y Tes s As a me hod o unde s and i TweeP o iles3 me he desi ed ea u es and design o end use s, we comple ed a second se o es s ( his can be seen in Appendix B). He e we made he sys em 4.2 Usabili y Tes s 51 a ailable and asked he pa icipan s o use i o ex ac in o ma ion om wee s, news and clus e s, e alua ing he ime and de ail o he esul . These es s we e made wi h people om JPN, bu un o una ely only 3 we e a ailable. While he pa icipan s we e using TweeP o iles3, we we e able o unde s and ha hey easily and e ec i ely ex ac ed in o ma ion om bo h wee s and all news ea u es. The abili y o apply di e en il e s o he da a was highly app ecia ed and used, being he con en il e ing he one ha go he mos a en ion. All da a shown on he map seemed o be use ul o he asks and he pla o m displaying he posi ion and de ails o he wee s, and no jus he densi y, was he one pa icipan s p e e ed, he eason being ha quali a i e in o ma ion is as impo an as quan i a i e. An example gi en o he usage o wee ’s de ails was he e o is a ack on Cha lie Hebdo. They said ha TweeP o iles3 would be pe ec o suppo an a icle ega ding wha happened, since hey could no only ga he in o ma ion om he news, bu also om wha people we e saying, gi ing he possibili y o inse ha in he a icle. Rega ding he same opic, Cha lie Hebdo, one aspec conside ed missing we e hash ags and im- ages. These wo ea u es a e emo ed om he analysis and since he bigges end om he a ack included images and hash ags, a lo o in o ma ion was los . Imp o ing he algo i hm o use his would be a big s ep o TweeP o iles3. One o he aspec conside ed impo an by he jou nalis s was he in eg a ion o Sapo pla o ms in TweeP o iles3. E en hough he news lis was mo e analysed han Máquina do Tempo, bo h we e acknowledge as a plus. One ea u e ha didn’ ge as much a en ion om hose using ou ool, was he di e en combina- ion o dimensions we included. Spa ial and empo al we e he dimensions ha pa icipan s mos used and analysed. This may be due o he ac ha hese p esen ed be e esul s and because wha in e es s jou nalis s he mos is wha people a e saying ega ding a opic o in e es and whe e hey do so. Wi h he sys em used, pa icipan s we e asked o comple e a o m, o us o unde s and how hey el abou he implemen ed ea u es and he gene al use o TweeP o iles3. This su ey was c ea ed based on [42] and people we e able o e alua e, om 1-5, di e en ea u es and aspec s o he sys em (1 being ha hey disag eed wi h he s a emen and 5 ha hey ully ag eed). A i s people e alua ed how simple he usage o he sys em was. All answe s poin o he ac ha TweeP o iles3 is a simple sys em o use (4, on ou scale). This was ou bigges conce n when designing he applica ion, o c ea e a simple sys em wi h all he desi ed in o ma ion. Ou espon- den s also ag eed (wi h he same e alua ion) ha hey we e able o comple e wo k e ec i ely using TweeP o iles3. Rega ding he usabili y o ou sys em, one pa icipan e alua ed wi h a 4, and one o hem wi h a 3, on ou 1-5 scale, how easy i was o unde s and. When i comes o he in o ma ion we p esen , all esponden s answe ed wi h 5 on how clea he in o ma ion was o ganized on he sc een, bu ega ding how easy i was o ind he in o ma ion 52 Resul s Figu e 4.13: Su ey ques ions ega ding ea u es om TweeP o iles3. hey we e sea ching, one pe son e alua ed wi h a 3, while o he s s ill e alua ed wi h a 5 (all on p e iously men ioned scale). One o he ques ion all pa icipan s came o ag eemen was ega ding he implemen ed ea u es. All said ha he mos impo an we e he map wi h all wee s/clus e s and he news and one pe son answe ed ha hey would emo e he lis o en i ies om Máquina do Tempo, since only known people could be sea ched. Tha same pe son said ha a ea u e which equi ed imp o emen we e he ime g aphs and ha he wo ds in he wo dcloud would need mo e wo k, in o de o be mo e de ailed. O he pa icipan s le hese ques ions in blank. When asking i hey conside ed he in eg a ion o he pla o m om Sapo, all pa icipan s an- swe ed posi i ely and also, all ag eed ha he news lis was mo e ele an han Máquina do Tempo. An aspec needed and desi ed o imp o e he usage o TweeP o iles3 is he possibili y o de ining opics o in e es , such as oo ball, poli ics, en e ainmen ... This would enable jou nalis o isu- alize da a ega ding only he heme hey would be in e es ed in, which would b ing mo e de ail o he ex ac ed in o ma ion. E alua ing all hese ea u es, asking i ou pla o m included e e y hing hey we e expec ing, he answe s we e 3 and 4 on ou scale. When inally asking he pa icipan s i hey we e sa is ied wi h he ou come o TweeP o iles3, he answe s we e posi i e, all being 4, which means ou sys em p o ed o be o he liking o all people who es ed i . E en hough he e a e some imp o emen s o be made and some ea u es o be added, and he numbe o jou nalis s we had o es ou applica ion was ai ly low, i was easy o unde s and 4.2 Usabili y Tes s 53 how all pa icipan s conside ed TweeP o iles3 as e , mo e e icien and de ailed han any o he adi ional me hod o ga he ing in o ma ion. Pla o ms like his a e seen as use ul and necessa y (which was answe ed on ou su ey wi h a 4, om all esponden s) o he jou nalism communi y, and o e all, TweeP o iles3 wen in he igh way, sa is ying all pa icipan s. 60 Appendix A - Ques ions made in he i s Su ey –Facebook / Myspace –Twi e / Mic oblogging –Mensagens de Tex o –Não usei –Ou a •Alguma ez usou uma ede social pa a di ulga in o mações ao público ( ela i amen e a jo nalismo/comunicação social)? –Facebook / Myspace –Twi e / Mic oblogging –Mensagens de Tex o –Não usei –Ou a •Alguma ez usou uma ede social pa a euni in o mações do público ( ela i amen e a jo - nalismo/comunicação social)? –Facebook / Myspace –Twi e / Mic oblogging –Mensagens de Tex o –Não usei –Ou a •Quais são os aspe os que mais alo iza em cada uma das e amen as que conhece, como uma e amen a de apoio ao jo nalismo/comunicação social? •Quais são os aspe os que menos alo iza em cada uma das e amen as que conhece, como uma e amen a de apoio ao jo nalismo/comunicação social? •Fe amen as de mic oblogging como o Twi e são ú eis, como e amen a de jo nalismo/- comunicação social, pa a a ecolha de in o mação –Conco do plenamen e –Conco do –Não conco do, nem disco do –Disco do –Disco do plenamen e •Fe amen as de mic oblogging como o Twi e são ú eis, como e amen a de jo nalismo/- comunicação social, de ido à ins an aneidade Appendix A - Ques ions made in he i s Su ey 61 –Conco do plenamen e –Conco do –Não conco do, nem disco do –Disco do –Disco do plenamen e •Fe amen as de mic oblogging como o Twi e são ú eis, como e amen a de jo nalismo/- comunicação social, de ido à in e ac i idade –Conco do plenamen e –Conco do –Não conco do, nem disco do –Disco do –Disco do plenamen e •Fe amen as de mic oblogging como o Twi e são ú eis, como e amen a de jo nalismo/- comunicação social, de ido à pe enidade –Conco do plenamen e –Conco do –Não conco do, nem disco do –Disco do –Disco do plenamen e •Fe amen as de mic oblogging como o Twi e são ú eis, como e amen a de jo nalismo/- comunicação social, de ido à mul imediação –Conco do plenamen e –Conco do –Não conco do, nem disco do –Disco do –Disco do plenamen e •Fe amen as de mic oblogging como o Twi e são ú eis, como e amen a de jo nalismo/- comunicação social, de ido à hipe ex ualidade –Conco do plenamen e –Conco do –Não conco do, nem disco do –Disco do 62 Appendix A - Ques ions made in he i s Su ey –Disco do plenamen e •Que ipos de in o mação geog á ica associada a edes sociais conside a ele an e pa a uma e amen a de jo nalismo/comunicação social? –Localização da pessoa que es á a ge a a in o mação –Localizações que são mencionadas ou ele an es pa a a no ícia (locais mencionados no Twee sob e um e en o, po exemplo) –Ou a •Que ipos de mapas associados a edes sociais conside a ú eis pa a o jo nalismo/comuni- cação social? –Mapas que mos em a localização da pessoa a pa ilha a in o mação –Mapas que mos em localizações mencionadas ou ele an es pa a a no ícia (locais mencionados no Twee sob e um e en o, po exemplo) –Mapas que mos em as duas an e io es –Ou a •Que e amen as e/ou ca ac e ís icas espe a e numa aplicação web in e a i a desenhada pa a inco po a edes sociais como on e de in o mação pa a apoio ao jo nalismo/comuni- cação social? –Tabelas –Mapas –G á icos de Tempo –Cha e / Tag Clouds –X / Y g á icos –P ocessamen o de ex o e e amen as de in o mação Fe amen as de Clus e ing –Fe amen as pa a Animação –Modelos p edi i os –G á icos de ede sociais –Coleções de Fo os/Vídeos –Ou a •Que ipo de in o mação espe a e numa aplicação web in e a i a desenhada pa a inco po a edes sociais como on e de in o mação pa a apoio ao jo nalismo/comunicação social? –Inciden es –Pessoas Appendix A - Ques ions made in he i s Su ey 63 –O ganizações –Sen imen os –Sequência de E en os ( imeline) –Resumos de Con e sação –In o mação Geog á ica –Ou a •In e ace Po a o e eja o design da in e aces que se segue. Rep esen a um esboço de design que se á desen ol ido pa a uma aplicação web in e a- i a que i ia apoia o pessoal de jo nalismo/comunicação social que p ecise de ex ai in- o mações ou pa ilha no ícias em edes sociais. Es e esboço usa o Twi e como on e de in o mação. Es e design ap esen a múl iplas componen es; (1) um mapa in e a i o que o e ece supo e básico e mos a os locais dos wee s assim como os clus e s iden i icados pela pla a o ma; (2) sis ema de isualização de clus e ing, capaz de mos a os wee s que o compõem, assim como pala as cha es de cada pe il; (3) uma e amen a pa a de inição dos pesos a ibuídos a cada uma das dimensões; (4) uma cloud de ags que suma iam pala as comuns nos wee s; (5) lis a de wee s baseado nas que ies ei as com a dimensão do empo, espaço e con eúdo. Se osse usa es a aplicação (ou uma semelhan e) como supo e ao jo nalismo/comunicação social, como usa ia? •Compa e o esboço com as seguin es aplicações (Demo 1) TweeP o iles - h p://you ube.com/wa ch? = pEZULbDHY4 Re wee Pa e n - h p://you ube.com/wa ch? =69JwdAmqqgc •Muda ia a espos a à pe gun a an e io ( ela i amen e ao esboço), endo em con a as apli- cações demons adas? •Que in o mações/ esul ados espe a se capaz de pa ilha a a és des a aplicação (ou uma semelhan e)? –Rela ó ios p é o ma ados –Link que inicia a aplicação & da a –Mapas pa a a imp essão –Cap u as de ec ã es á icas –Pequenos clips de ídeo –Ou a 64 Appendix A - Ques ions made in he i s Su ey Figu e A.1: TweeP o iles3 mockup. Appendix B Appendix B - Ques ions made in he second Su ey •Qual é o seu nome? •Qual é a sua idade? •Quais as suas habili ações académicas? •Qual é a sua unção p o issional a ualmen e? •Alguma ez abalhou numa á ea elacionada com o jo nalismo/comunicação social? •Se sim, quan o empo? •É um sis ema simples –1 –2 –3 –4 –5 •Consigo comple a o meu abalho e icazmen e com es e sis ema –1 –2 –3 –4 –5 •Foi ácil de en ende a usabilidade do sis ema 65 66 Appendix B - Ques ions made in he second Su ey –1 –2 –3 –4 –5 •Foi ácil encon a a in o mação que p ocu a a –1 –2 –3 –4 –5 •A o ganização da in o mação no ec ã é cla a –1 –2 –3 –4 –5 •O sis ema con ém as unções e e amen as que es a a à espe a –1 –2 –3 –4 –5 •Es ou sa is ei o com o TweP o iles3 –1 –2 –3 –4 –5 •Que ca ac e ís icas do TP3 achou mais impo an es? Appendix B - Ques ions made in he second Su ey 67 –Mapa com wee s/clus e s –Wo dcloud –G á icos –No ícias –Máquina do Tempo •Re i a a alguma e amen a da aplicação? Qual? –Mapa com wee s/clus e s –Wo dcloud –G á icos –No ícias –Máquina do Tempo •Alguma ca ac e ís ica que conside a ele an e mas que necessi a de melho ia? –Mapa com wee s/clus e s –Wo dcloud –G á icos –No ícias –Máquina do Tempo •Conside a as e amen as de ligação ao Sapo uma mais alia? –Sim –Não •Qual oi mais ele an e? –No ícias –Máquina do Tempo •Conside a a in o mação (pala as) con ida nos clus e s ele an e? –Sim –Não –P ecisa de mais de alhe •Conside a a possibilidade de escolha de emas ( u ebol, polí ica, en e enimen o...) uma ca a e ís ica impo an e pa a o u u o? –Sim 68 Appendix B - Ques ions made in he second Su ey –Não •Compa e a ecolha de dados no TP3 com um mé odo adicional. Qual é mais ápido, e icien e ou de alhado? •Vê necessidade/u ilidade numa e amen a como o TweeP o iles3? –Sim –Não •No ge al, es ou sa is ei o com o esul ado inal –1 –2 –3 –4 –5 B.1 Planning: Ini ial Final Ou ini ial planning o he disse a ion didn’ su e majo changes h ough ou he whole de elop- men p ocess. Some p oblems we e encoun e ed ega ding TweeP o iles2 and SocialBus, leading o some delays and o u n he comple ion o he social dimension in o a seconda y objec i e, due o lack o ime. We also added a second se o in e iews, in o de o ully unde s and he usage o he de eloped sys em. O e all, he schedule was me . Figu e B.1: Ini ial Gan Diag am. B.1 Planning: Ini ial Final 69 Figu e B.2: Final Gan Diag am.