Geotagged tweets to inform a spatial interaction model: a case study of museums
This paper explores the potential of volunteered geographical information from social media for informing geographical models of behavior, based on a case study of museums in Yorkshire, UK. A spatial interaction model of visitors to 15 museums from 1…
Authors: Robin Lovelace, Nick Malleson, Kirk Harl
Ge o t a gg e d t w e e ts t o in form a s p a ti a l in te rac t io n m odel: a c a s e s tu d y o f m use u m s Ro b i n L o v el a c e , N i c k M a l l es o n , K i rk H ar l a n d an d M ar k B i rk i n School of G eography, Univ er si t y of L eeds , Leeds , U K R .L ov el ace@ l eeds .ac. u k Ge o tag g e d twe e ts to in form a s p a ti a l in te rac t io n m odel: a c a s e s tu d y o f m use u m s T h i s paper ex pl ore s t h e pot en t i al of v ol u n t ee red g eog raph i ca l i n f or m a t i o n f rom soci al m edi a f or i n f or ming g eog raph i ca l m odel s of beh av i or , based on a ca se st u dy of m u s eu m s i n Yor ksh i re, U K . A spa ti al i n t era ct i o n m odel of v i si t ors t o 15 m u se u m s f r om 179 adm i n i st r at ive z on es i s con st ru ct ed t o te st t h i s poten t i al . T h e m ai n i n pu t dat aset com pri ses g eo - t ag g ed m ess ag es h ar v es te d u si n g th e T wi tt er St ream i ng A ppl i cat i o n Prog r am mi ng I n t er f ace ( A P I) , f i l ter ed, an al y z ed an d ag g r eg at ed to al l ow di rec t com pari son w i th t h e m odel ‟s ou tpu t . Com par i s on bet w een m odel ou t pu t a n d tw eet i n f or m atio n al l ow ed t h e cal i br at i o n of m odel par am et ers to opt i m i z e th e f i t bet w een f l ow s t o m u se u ms i n f err ed f rom tw ee ts and f l ow m at ri ces g en er at ed by t h e spat i al i n t er act i o n m odel . We con cl u de th at v ol u n t e ered g eog r aph i c i n f or m atio n f r om s oci al m edi a si tes h av e g r eat pote n t i al f or i n f or m ing g eog raphi ca l m odel s of beh av i or , es peci al l y i f t h e v ol u m e of g eo - t ag g ed soc i al m edi a m ess ag es con t i n u es to i n cr ease . H ow ev er, we cau ti on t h at v ol u n t e ered g eog r aph i ca l i n f orm a t i o n f rom soci al m edi a h as som e m aj or l i m i ta t i o ns so sh ou l d be u sed on l y as a s u ppl em ent to m ore con si st en t dat a sou r ces or w h en of f i c i a l dat aset s ar e u n av ai lab l e. Ke ywo rd s: Sp ati a l i n tera c ti o n, vo l u n te ere d d a ta , soci a l m e d i a , V G I In tr od u cti o n Mu ch h as been w ri tt en abou t th e pot en ti al of v ol u n t eer ed g eog raph i c a l i n f or m atio n ( VG I) f or i n f orm i ng g eog raph i c res ear ch proj ect s (1) . Ge o- coded soci al m edi a dat a pr ov i ded by a di f f u s e n et w ork of se l f sel ect ed u s ers pres en ts a ri ch y et ch al l engi ng data sou r ce (2) th at cou l d, i f f i l t ere d an d i n t erpr et ed w i t h su f f i c ie n t care , prov i de u sef u l i n pu t s i n to m an y are as of g eog r aph i c re sear ch (3) . In th i s pape r w e dem on st rat e a n d di scu ss th e poten t i al of VG I as an i n pu t i n t o g eo- sci en ce u si n g a cas e st u dy of a spa ti al i n t er act i o n m odel ( SIM) cal i br at ed u si n g h ar v est ed T w i t t er da ta. SIM s h av e a l on g h i s tor y i n academ i c an d com m erci a l appl i cat i o n s of g eo- sci en ce. T h ei r u t i l i t y i s dem on st ra te d by pers i st en t u se i n a ra n g e of f i el ds i n cl u d i ng r eta i l pl an n i ng an d st or e l ocat i on opti m i z at io n (4, 5) , tr an spor t pl an n ing (6) an d th e i n v es t i g atio n of m i g rati o n (7, 8) . T o i n tr odu ce t h e SIM con cepts an d i l l u st rat e t h e ki n d of probl em t h at th e t ech n i qu e ta ckl es, l et u s con si der a con cre te case st u dy i n som e det ai l r at h er t h an t o at t em pt an ex h au st iv e re v i ew on th e l ar g e body of l i t erat ur e on th e su bj ect . A r ecen t st u dy i n v est ig a t ing f l ow s of ag g reg at es – rock- base d m at er i al u s ed i n t h e con st ru ct i o n i n du st ry – by Z u o et al . (9) ser v es th i s pu rpos e w el l . Spat i al i nt erac t i on m odel s SIM s are m ost f requ en t l y appl i ed to es ti m ate t h e f l ow ra te (T ij ) bet w een a n u m ber of or i g i n s (i ) an d dest i n at i o n s (j ) , as a f u n ct i o n of t h e di s tan ce betw een th em an d th ei r ch ar act eri st i c s. T h e ou tpu t i s u su al ly a f l ow m at ri x w i t h row n am es corr es pon di ng t o or i g i n s an d col u m n h eadi n g s t o dest i n at i o n . T h e SIM i s u se d t o est i m at e t h e cel l v al u es of t h e f l ow m at ri x . T h e cl ass i c v er si on of a SIM , w i t h th e ou tpu t bei n g t h e ori g i n - des ti n at i o n f l ow m at ri x T , i s pres en te d i n equ at i on 1 bel ow : (1) w h ere O i an d D j r epres en t each or i g i n an d dest i n ati o n (or , i n m or e com pl ex m odel s, som e f u n ct i o n of t h ei r ch arac ter i st i cs ); A i an d B j are w ei g h t s as soci at ed w i th t h e or i g i ns an d dest i n at i o ns re spect i v ely ; an d d ij i s t h e di st an ce or t ran s port cost s t h at separ at e t hem i n g eog r aph ica l spac e. A s Wi l son ( 10) dem on st r at es, t h e tr eat m en t of th e we i g h ts (A i an d B j ) i s of cr u ci al i m port an ce i n th e m odel i ng proce ss. Wi th ou t re st ri ct i on on ei th er of t h es e t erm s ( or w h en t h ey are a bsen t ) t h e r esu l ti ng m odel i s con si dered „u n con st ra i n ed ‟ . A l ter n at iv e l y t h e w ei g h t s can be u se d as „ bal an ci ng f act ors ‟ t o m ai n ta in con si st en cy i n ei th er t h e ori g i n tot al s or dest i n at i o n tot al s or bot h si m ul t a n eo u s l y , i n t h e case of a dou bl y con st r ai n ed m odel . I n t h e case of t h e paper on f l ow s of roc ky ag g r eg ate s f or t h e con st r u ct i o n i n du st r y ( 9), t h e ori g i n s ar e t h e qu ar ri es f rom w h i ch m at eri a l s are m i n ed an d th e dest i n at i o n s are t h e di st ri ct s to w h i ch t h e m ate ri a l s are de l i v ered. B al an ci ng f ac tor s ar e em pl oy ed t o con st rai n t o kn own ag g r eg at e f l ow s f rom each qu arr y an d to eac h di s tr i ct. H en ce th e m odel wa s speci f i ed w i t h th e f ol l ow i ng f u n ct i o n a l f or m ( 9): (2) (2A ) (2B ) w h ere A i an d B j ar e th e bal an cing f ac tor s, O i an d Dj ori g i n s an d dest i n at i o n s an d β i s t h e di st an ce decay param et er , to be cal i bra te d. T h e bet a ( β) v al u e i s cru ci al i n SIM s as i t det er m i n es th e di st an ce de cay ef f ect betw een ori g in s an d dest i n ati o n s an d su bse qu en t l y t h e ex ten t t o w h i ch f l ow s bet w een n ei g h bor i ng ori g i n s an d dest i n at i o ns ar e f av or ed ov er l on g er di st an ce f l ow s. M u ch res earc h h as f ocu s ed on th e v ar i abi l ity of β v al u es an d th e f u n ct i o n a l f or m of di st an ce - decay f ( d ij ) ( 11, 12). I n t h i s pape r w e al so f ol l ow Wi l son (10, 13) i n ch oosi n g an ex pon en t i a l decay : (3) T h e des ti n at i o n z on es ar e re pres en te d as poi n ts , al l ow ing pre ci s e di st an ces (d) t o be cal cu l ate d. Fi n al ref i n e m e n t s to th e SIM i n cl u ded addi t i on a l f act or s su ch as t h e pr ox i mi t y of tr an spor t l i n ks to ea ch qu arr y , on ce com pl et e par am et er v al u es w er e es ti m ate d th rou g h cal i brat ion (9) . C al i br at i o n i n v olv es i t erat i v e l y est i m ating para m et er s t o opti m i z e th e f i t bet w een obser v ed an d m odel ed f l ow s; i n th e m odel pre sen t ed h ere T w i tt er dat a i s u sed to c al i bra te β i n equ at i on 3, an d ex pl ore ot h er para m et er s t h at cou l d be opt i m i z ed. I n th e m odel of m u seu m v i s i t s pres en t ed h ere, th e con f i gu r a t i o n of th e SIM i s sl i g ht l y di f f er e n t t o th at pres en ted by Z u o et al . in (9) : th e ori g i n s ar e di f f u se adm i n i st r at i ve z on es an d t h e dest i n at i o n s are a l i m i ted n u m ber o f u n ev enl y di st ri bu t ed poi n t s – t h e m u seu m s – l oca ted i n an d ar ou n d t h e E n g li s h ci ti es of L eeds an d B radf ord. A si m pl if i ed re pres en ta ti o n of th i s m odel set- u p, f or 10 ori g i n z on es an d 2 m u se u m s, bas ed on re al l ocat i on s, i s i l l us tr a te d i n Fi g u re 1 . Fi g u re 1. Si m pli fi ed i l l u st rat i o n of an u n con st r a i n ed spat i al i n t era cti o n m ode ( SIM) , bas ed on 10 w ard- l ev el adm i n is tr a t i v e z on es an d t wo w el l - kn ow n m u s eu m s i n L eeds . T h e f l ow of v i si tor s t o each m u seum i s i l l u st rat ed by t h e th i ckn ess of each st ra i g h t l i n e . G eot agged soci al m edi a t o i nform a s pat i al i nt er act i on m odel I n th i s pape r w e sh ow h ow th e S I M st ru ct u re des cri bed abov e can be appl i ed, t o ex pl ore m ov em ent patt er n s of peopl e v i s i t i ng m u se um s. T h e key i n n ov at i o n i s th e u se of VGI , t aken f r om T wi tt er , to i n f or m th e m odel . Met h ods an d ex t er n al da ta t o ef f ect iv e l y cal i brat e SIM s h av e been a l on g - ru nni ng con cer n f or res ear ch ers ( 5, 14, 15). Cri t i cal l y , w e dem on st r at e t h at soci al m edi a dat a can ai d th e con st ru ct i o n an d cal i br ati o n of SIMs , en abl i ng th e i n cl u s i o n of v ari abl es n ot av ai l ab l e f rom tr adi ti on a l sou rce s. Met h ods of h arn es si ng g eo- ref er en ced soci al m edi a dat a f or f l ow m odel cal i br at i on are al so dem on st rat ed, h i g hl i gh t in g t h e n eed f or i n pu t dat a t h at i s su f fi c i e nt ly l arg e an d ri ch f or t h e pu r pose. We dev el oped th e m et h ods i n R , a f r ee an d open sou r ce st at i st i ca l pr og r am mi ng l an g u a g e t h at h as al r eady been u sed f or cal i br at ing SIM s (16) . Wh i l e SI Ms h av e occas i on al ly been appl i ed t o th e rel ate d ph en om en a of j ou rn ey - to - w ork f l ow s ( 17, 18) an d t ou r i s t t r av el (19, 20) , t h es e pers on al t rav el m odel s t e n d t o as s um e t h at pa tt e r n s ar e re gul a r a n d c y c li ca l – i n a s e n se a l o ng - te r m eq ui l i br ium . T h e real i ty i s m u ch m ore v ar i ed – rec en t res ear ch u si n g GPS t ra cki n g dev i ces sh ow s t h at h u m an beh av i or i s com pl i cat ed, w i t h seem i n gl y st och ast i c v ar i abi l ity i n obse rv ed t i m i ng s an d spat i al di st r i bu t io n s of ev er y day act i v i t y pat te rn s ( 21, 22). On e cou l d al s o spe cu l at e t h at , w i th i n cr ease d com m u nic a t i o n bet w een peopl e an d pote n t i al f or f r ag m enta t i o n of t rav el pl an n i ng sou r ces du r i n g t h e on- g oi n g di g i ta l rev ol u ti o n ( 23, 24), t h e ten den cy i s tow ar ds g re ate r com pl exi t y . T ime - ser i e s datas et s are g en era l l y t oo sc arce t o tes t su ch h y pot h eses at pre sen t bu t th e g en er al poi n t , th at per son al tr av el i s of t en n ot re g u l ar t em por al ly or pr edi cta bl e s pat i al l y i s w i del y acce pted (25) . Su ch u n predi ct ab il i t y an d „m es si n e ss ‟ are no t w el l ca ptu r ed i n m u ch t ra n sport r esear ch (s ee (26) f or a di scu s si o n of th i s i ss u e f rom th e pers pect i v e of tr af f i c saf et y m odel i ng), as i l l ust ra ted by t h e om i ss i on of th e f ol l ow i ng ph en om en a f rom m ost con tem porar y t ran s port m odel s: Mu l ti - pu r pos e jou rn ey s, t h e ten den cy t o ch ai n t og et h er t r i ps of di f f er e nt t y pes . T h e tr i p ch ai n h om e- sch ool - w ork- sh op- w o rk- gym - bar- h o me , f or ex am pl e, m ay be r epre sen t ed as a s i m pl e h om e- w ork tr i p. Seas on al , w eekl y an d di u r n a l v ari at i on i n t h e ti m i ng an d f r equ en cy of tr i p s – f or ex am pl e du e t o h ol i day s, „f l exi - t im e ‟ w ork con t ract s , an d ch an g i ng con su m er h abi ts . Up - to - date i n f or m atio n abou t t h e st at e of t h e t rav el n etw ork du e to i m pr ov ed i n t er n et com m u ni c at i o ns i n com bi n at i o n wi t h th e ra pi d u pta ke of sm art ph on es i n cr eas ingl y i n f lue n ces th e t i m i ng an d r ou t e ch oi ce of t ri ps. R eal - t i m e tr af f i c i n f or m atio n f r om G oog l e or di r ect l y to i n - car sat el l it e n av i g at i o n sy st em s an d t h e cu rr en t st at e of bi ke h i re h u bs are ex am pl es of f act ors th at cou l d i n f l u e n ce t i m i ng s, m ode an d rou t e ch oi ce. Th e a bo v e po int s a r e esp ec i a l ly r e l e va n t t o i rr e gul ar tr ip de m a n ds (e. g . s h opp i ng, g y m , l ei su re) : t h es e are l ess predi ct abl e th an re g u l ar t r i ps su ch as com m u t ing , an d are t h eref ore ch al l engi ng t o m odel . T h e st u dy of m u seu m v i si ts i n t h i s pa per i s an ex cel l ent ex am pl e of su ch i rr eg ul ar t ri p - m aking beh av i or. O n e di f fi c ul t y ass oci at ed w i t h an al y zing t h e ex ten t of v ari abi l it y i n tr av el pat te rn s i s th at th e dat a h as been di f f i c ul t an d ex pen si v e to ac q u i r e - m os t tr av el di ar y an d ot h er of f i c i a l dat a col l ec ti o n te ch n i ques t en d to as su m e reg u l ar t r i ps an d m as k v ari abi l it y . T h i s i s st art i n g t o ch an g e as par t of t h e „B i g Da ta re v ol u t i o n ‟ : cr ow d - sou rce d, soc i al m edi a an d pas si v el y h ar v est ed dat ase ts ar e bei n g col l ect ed on a v ast scal e. T h e academ i c se ct or h as been sl ow t o take adv an t ag e of th es e n ew dat as ets ( 27) despi te wi despr ead ackn ow l edgm e nt of t h ei r pot en ti al f or dev el opi ng ou r u n ders tan d i ng of tr av el pat te rn s an d beh av i ors. O n e barr i er t o u ptake i s data qu al i t y: tw ee ts , f or ex am pl e, con t ai n a m ax i m um of on l y 140 ch ar act ers an d are n ot con st ra i n ed i n an y oth er w ay , posi n g m aj or ch al l eng es f or i n t er pret at i o n . A n oth er obst acl e i s t h at soci al m edi a u ser s ar e a s el f - sel ect ing sam pl e, u n l i ke ly to be r epre sen t at iv e of t h e w i der popu l at i o n . O n e cou l d arg u e t h at t h e m ost prol i f ic on l i n e com m u ni ca tor s are w h ol l y unre prese nt at iv e , con st i tu t i ng a sm al l m i n or i t y of t h e popu l at i on w i t h an u n u su a l at t ach m en t to di g i ta l dev i ce – th e g eo- ref ere n ced t we et s u sed i n th i s paper , f or ex am pl e, con st i tu t e on l y arou n d 2% of T wi t te r i n f or m at i o n ou tpu t an d can n ot be as su m ed to be r epres en ta ti v e of al l t w eet s, l et al on e of soc i et y as a w h ol e (2) . A cadem i cs h av e ri g h t l y been cau t i ou s of t h es e i s su es, bu t we arg u e t h at m u ch m or e u se can be m ade of th e v ast st or es of v ol u n t eere d g eog raph i ca l i n f orm at i o n th at are av ai l ab l e , espe ci al ly i n th e dat a depr i v ed ar ea of m odel i ng com pl ex i rr eg u l a r t rav el pat t ern s. Resear ch over vi ew T h i s paper ex am i n es w ay s of tac kl i n g t h i s r esea rch g ap wi t h an ex pl or at ory ex am pl e of m u s eu m v i si t s an d com m u ni cat i o n s abou t m u seu m s, u t i li zi ng T w i tt er dat a as a sou r ce of i n f or m atio n abou t att i tu des t owa rds an d spat i al patt er n s of m u seum v i s i t s. T h e se soc i al m edi a dat a w er e capt u r ed by h arv es ti ng tw ee ts i n th e case st u dy ar ea su r rou n di ng L ee ds an d B ra df ord, U K , ov er an 18 m on t h peri od . T h e g eog raph i c dat a on m u s eum s w as col l ec te d f rom O pen Str eet Map. Th e f ol l ow i ng sec ti on of th e paper descr i bes th e i n p ut da t a in mor e d et a i l. Va r io u s fi l t er s w er e u sed to e x t ra ct s u bsets o f „ m u se um twe e t s‟ f r om t h e paren t dat a set an d t h e proper t i es of t h i s su bs et ar e descr i bed an d an al y z ed. T h e st eps w h i ch h av e been take n to i n f er t h e res i den ce of t h e „t w eet er ‟, t h e l ocat i on of t h e m u seu m or ex h i bit i o n an d ot h er rel ev an t pl aces e.g . t h e sou r ce of t h e m es sag e ar e al so pre sen t ed. Sect i on 3 ou t l i n e s h ow a si m pl e u n con s tr a in ed S IM w as u sed t o est i m at e t h e f l ow s bet w een adm i n is t ra t iv e z on es an d m u se u ms an d t h e st eps t aken to s el ect i n i t i a l m odel par am et ers du ri n g cal i br at i o n . R ef i n e m e n t s to t h e m odel con s i der v ari at i o n s i n t h e att r act i v en es s of di f f er e n t ex h i bi ts , v i si tor dem og r aph i c s an d tr i p - m aki ng pr open si ti es. T h e f i n a l sec ti on rev i ew s an d di s cu sses t h e f i n ding s an d pre sen t s con cl u ding com m en t s. A l th ou gh t h i s r esea rch pr ojec t en g ag es di re ct l y wi th t h e con cept of „ bi g dat a‟ t h e ex t rac t an al y se d h ere i s act u al l y rat h er l i mi t ed ; av en u es f or ex t en di ng th e sam pl e s i z e ar e bei n g con si dered. Pot en ti al appl i cat i o ns f or a m odel of th i s ki n d are w i despr ead , f or ex am pl e i t cou l d be u sed to ex am i n e th e ef f ect i v e n e ss of adv ert i si ng f or speci f i c sh ow s, or to t rack pu bl i c s en ti m e n t tow ar ds an d appre ci ati on of di v ers e ex h i bi t i o ns am on gst di f f er e n t au di en ces. M ore i n t er est i ng f r om an academ i c v i ew perh aps i s h ow t h i s beg i n s to i l lu strat e n ov el i n ter act i o n patt er n s i n a ci ty , w h i ch are n ot on l y i n t er est ing i n t h ei r ow n r i g h t bu t per h aps ev en m ore so as i n di vi d ua l el em en t s i n an i n cr easi ngl y com pl ex w eb of dai l y u rba n m ov em en t s. D ata an d M e th od s : h ar ve s ti n g an d fi l te r i n g t h e tw e e ts The mus eum dat a T h e m u seu m s i n t h e cas e st u dy ar ea w er e i den t if i ed u si n g a spat i al l y bou n ded sear ch of O pen St ree t Map ( O S M) dat a. T h e en ti re ty of U K O SM data w as f i rs t down l oad i n t h e com pres sed Prot ocol bu f fer B i n ary Form at (.pbf ). T o re du ce pr ocess i n g ti m e, th i s f i l e w as cl i pped t o t h e st u dy ar ea. 1 A f t er tr an sf err i ng t h e dat a i n t o a spat i al dat aba se an d t h en a G IS, t h e at t ri bu te t abl e w as an al y z ed. . B ot h „ n am e‟ an d „ tou r i st ‟ col u m ns w ere u sed t o sea rch f or t h e key w ord „ m u seu m‟ , w i t h ou t cas e sen s i t ivi t y . A l l n am ed m u seu m s w er e al so cl as si fi ed as m u seu m s, so t h e tou r i s t at t ri bu t e v ar i abl e w as u sed f or i den ti f i ca t i o n . 2 I n tot al t h er e w er e 7 poi n t obj ect s an d 13 pol y g on obje cts tag g ed as " m u seum " i n th e col u m n " t ou ri s m ". H ow ev er , on e of th e poi n t s 3 du pl i cat ed a pre- ex i st i ng pol y g on of t h e sam e bu i l di ng an d so w as de l et ed. B radf ord In du st ri a l Mu se u m , i n th e n or th - eas te rn ou t ski rt s of th e ci t y , i s com posed of m u l t i p le bu i l di ng s an d w as re pl i cat ed 5 t i m es. On ce t h es e i s su es h ad been res ol v ed, t h e t ot al n u m ber of m u se um s i n th e st u dy ar ea w as 15. T h ese we re al l con v er te d i n t o poi n t data f or con si st en c y (f ig u re 2 ). 1 To c l i p th e .pb f fi l e th e com m a nd - l i n e OSM c o nv ers i o n to ol „osm con v ert‟ wa s u se d . Th e fo l l o wi ng co m m a n d wa s a p pl i e d to th e n a ti o n w i d e d a ta, re sul ti n g i n a 5 -fo l d re d ucti o n i n fi l es i z e: o sm c o nv ert l e e dstw.o s m .pb f > l e e d s tw.osm. 2 Th e a ttri bu tes o f OSM e l e m e n ts, i ncl u d i ng „n a m e ‟ a n d „to u ri s m ‟ a re d e s c ri b e d h e re : h ttp : // w i k i .op enstree tm ap.o rg /wi k i /Map _ Fea tu re s 3 Th i s po i n t wa s Bo l l i n g Ha l l Mu se u m , on th e So u th er n ou tsk i rts o f Bra d fo rd. Not e i n fi gu re 2 th at th i s m u s e u m i s m i sspel t i n th e m a p , u n derl y i n g d a ta qu al i ty i ssues a ss o c i a ted wi th OSM d a ta. Fi g u re 2. L oca ti on s of t h e 15 m u seu m s u sed i n th e case st u dy . B asem ap: G oog l e. T h e T w i tt er dat a w er e col l ect ed u si n g th e T w i t t er St re am i ng A PI. T h i s i s a se rv i ce t h at prov i des l i mi t ed acces s to pu bl i c m essa g es post ed t o T wi t te r. D at a w ere col l ec te d du r i n g 445 day s bet we en 2011 - 06- 22 an d 2012- 09- 09. T h e f u l l dat aset con si st ed of 992,423 t w eet s ori g i n a t ing i n t h e L eeds ar ea . E ach r ecord i n t h e dat aset r epres en ts on e t w eet an d i n cl u des a t i m es ta m p f or th e g en era ti o n t i m e, a u ser i d al l ow i ng t w eet s f r om t h e sam e acc ou n t to be l i n ked t og et h er an d g eog raph i ca l coor di n at es of th e l ocat i on wh er e t h e tw ee t ori g i n at ed. O n l y m ess ag es t h at h av e as soci at ed g eog raph i ca l coordi n at es h av e been i n cl u ded i n th e an al y si s – r ecen t es ti m ate s su g g es t t h at th i s r epre sen t s 1 - 2% of al l m es sag es ( 2). Su ch m ess ag es a re com m on l y cr eat ed u si n g m obi l e dev i ces by u ser s wh o h av e ex pl i ci t l y opte d t o pu bl i sh t h ei r pres en t l ocat i on . U si n g th e data m an i pula t i o n f u n ct i o ns i n t h e My SQ L dat abase m an ag em e nt sy st em a cu rs ory an al y si s of th e col l ect ed dat aset w as u n der t aken . Most u ser accou n ts sh ow ed tw eet i n g act i v i t y of l es s t h an 1,000 du ri n g th e per i od of st u dy . H owe v er, a sm al l n u m ber of accou n t s, 274, h ad act i v i t y m u ch h i g h er t h an 1,000 tw eet s . Fu rt h er i n v es t i g atio n re v ea l ed 12 ac cou n t s t h at we re bot h prol i fi c u ser s an d g eog ra ph i ca lly st at i c. Fu r th er ex am i n a t i o n of th e l i n guis t i c st r u ct u re w i t h i n th e tw eet ed te x t i den t i fi ed r epet i t i v e pat t ern s repr esen t i ng w eat h er s ta ti on s, car sal es prom oti on s, tr af f i c wa rn i n g s an d adv ert i si ng f or on l i n e m ag az in es . T h e 12 acc ou n t s i ssu i ng au t om at ed i n f or m a t i o n w er e r em ov ed f rom s u bs equ en t an al y s i s. T h e rem ai ning dat aset com posed 958,339 t w eet s ov er 27,999 act i v e u ser accou n t s. T h e tex t f r om eac h tw eet wa s decom posed i n t o a se ri es of w ords . T h e r u l e u sed t o i den t i fy wor ds i n t h i s an al y si s i s a se t of ch ar act ers en capsu l at ed by ei th er bl an k spa ce or a com bi n ati o n of bl an k spa ce an d a pu n ct u at i o n ch ar act er su ch as a f u l l st op, com m a, ex cl am at i o n m ar k, qu est i on m ar k, col on or s em i - col on . Words con si s ti ng of on e or t w o ch ar act ers w ere ex cl u ded f r om f u r th er an al y si s l eav ing 11,505,7 19 w ords f or m ing th e 958,339 t w eet s ( 28). Fi l t erin g t he t w eet s T h e n ex t st ag e i n t h e an al y s i s wa s to f i l t er t h e tw eet s t o i den t if y th ose rel at i ng to m u s eu m v i si t s al l ow ing t h e i n v est i g a t i o n of t h e spa ti al beh av i or of m u seum v i si t ors . T w o m ai n opt i on s f or f i l t er i ng a pr edef i n ed st u dy area ar e to appl y a spat ial or a se m ant ic f i l ter . T h e f or m er i s u su al ly si m pl y i m pl e m e n t ed as a „ cl i p‟ f u n ct i o n to r em ov e an y i n f or m atio n ou t si de t h e ar eas of i n ter es t, bu t can al s o be appl i ed by al ter i n g th e pr obabi l i ty of sel ect i on or r an k of obser v at i on s based on th ei r prox i mi t y to th e l oca ti on of i n t er est (29) . Sem an ti c f i l t er s i n t erpr et th e u ser ‟s m ean i ng f r om ch ar act er st ri n g s w i th i n th e t ex t of t h ei r m es sag e. T h i s t y pe of t ex tu al f i l t er can al so be con st r u cte d at di f f er e n t l ev el s of com pl ex i t y , f r om si m pl e key wor d sear ch es - al l T we ets con ta i ni ng “ m u s eum” , f or ex am pl e - to m ore com pl ex t ech n i qu e s (30) . Mor e soph i st i cat ed sea rch es i n cl u de „f u z zy m atc h i ng ‟ t o accou n t f or m i sspe ll i ng s or sy n on ym s (31) an d t h e cl ass i f i c at i o n of t w eet con t en t i n t o sem an t i c cat eg or i es to ex t ract g r eat er m ean i ng f r om t h e com bi n at i o n of w ords u sed ( 32). For ex am pl e by al l ocat i ng f ac ets , or m ean i ng cat eg or i es , u si n g t ech n i qu es of N at u ral L an g u ag e Proces si n g (NL P) (33) . Sem an ti c se arch es can g o f ar bey on d si m pl e key w ord sear ch es, an d i t h as been f ou n d t h at t h i s can y i el d ben ef i ts t o u ser s l ooki n g f or con tex t- speci fi c i n f orm at i o n (34) . B el ow w e i l l u s tr at e m eth ods to i m pl e m e n t bot h t y pes of se arch to s el ec t t w eet s abou t m u seu m s. T h e spatial f il t er i n t h i s con t ex t wou l d sel ect on l y t h ose t w eet s re corde d w i t h in or v er y cl os e to kn ow n m u se u m s . B el ow we u s e t h e R oy al A rm ou ri es, a n at i on a l m u s eu m of h i s tor y an d ar m am en t s th at m ov ed to L eeds i n 1996 (35) , to i l lu strat e th e m eth od. T h e f i rst s tag e wa s t o i den ti f y i ts g eog r aph i ca l coor di n at es, u si n g an on l i n e se arch . We u sed th e sear ch en g i n e „w w w .g eon am e s.or g ‟, al th ou gh an y on e ou t of a n u m ber of si te s y i el ded th e sa m e l at / l ong coordi n at es f or i ts cen t roi d: 50.8607 / - 1.1389. A st rai g h t f or w ard spat i al f i l t er w ou l d s el ec t al l t w eet s w i t h a cer t ai n di st an ce ( e.g . 200 m ) of th i s l ocat i on , t o capt u r e al l t w eet s se n t i n an d di r ect su rr ou n ding th e bu i l ding . T h i s i s cl ear l y an i n adequ at e m et h od becau se m u seum s are r ar el y ci rc u l ar bu i l di ng s an d t h ei r si z e ch an g es g r eat l y f r om on e to t h e n ex t ( see f i g ur e 3). A f l oor- plan of th e m u seum s u n der i n v est i g a t i o n i s requ i re d so a m ore accu r at e spat i al f i l te r can be cre at ed. T h i s data is av ai l abl e f rom a v ari et y of l ocal sou rce s i n cl u d i ng t h e l ocal Cou n ci l or t h e E st ate Man ag er at t h e m u seu m i t sel f . B u i l d ing f l oor pl an pol y g on s are al s o av ai l ab l e at th e n at i on a l l ev el f or t h e UK f r om t h e O rdn an ce S u rv ey ‟ s Ma st erM ap produ ct , h ow ev er t h i s dat ase t i s l ar g e an d requ i res a l i cen se f or acces s m aki n g acces s probl em ati c. 4 For 4 Mas terMap d a ta ca n b e d own l o a d e d for fre e un d er a n ac a dem i c l i c e n s e, bu t o n l y i n 1 0 by 1 0 k m c hu n k s . Th e si z e of th e se 1 00 k m 2 c h un k s i s l a rg e , 16 0 Mb fo r th e a re a d i re c tl y su rrou n d i n g Lee ds. r eprodu ci bi li t y , ease of acce ss an d ( al bei t pat ch y ) i n ter n at i o n a l cov er ag e, w e deci ded t o u se O pen St ree t Map (OSM), a crow d sou rc ed w orl dw i de m ap f rom w h i ch t h e raw dat a i s pu bl i cl y av ai l ab l e. T o acce ss O SM v ect or dat a, an d n ot ju s t th e ra st er dat a t h at i s di spl ay ed on t h e O SM w ebsi te, w e u s ed osm ar, an R packag e th at con n ect s t o O SM‟s A PI. B as ed on t h e af orem en t i o n ed g ri d re f er en ce, t h e f ol l ow i ng code w as u sed t o down l oad vec tor dat a con t ai ning th e f l oor- pl an of th e R oy al A r m ou ri es Mu se u m an d sa v e i t as an obj ect cal l ed „ar m ‟: b b < - ce n t e r _ b box( - 1 . 5 3 2 3 , 5 3 . 7 9 1 9, 6 0 , 6 0 ) a r m < - g e t _ o sm( b b , s o u r c e = s r c ) T h i s r esu l ted i n t h e pol y g on sh ow n i n f i g u re 3. A f t er th e pol y g on w as con v er te d i n t o a con v en ti o n a l R spat i al objec t an d i t s coor di n at es we re t ran sf or m ed i n t o a l oca l coor di n at e sy s tem (O SG B 1936, E P SG code 27700) , t h e ar ea w as cal cu l at ed an d th e t ru e cen tr oi d of th e bu i l d ing i n t h e n ew coor di n at e sy st em w as re corde d. 99 t we et s w ere r ecorde d f r om w i th i n a 10 m et r e bu f f er of th e m u seu m ‟s f l oo r- pl an . O f th es e on l y on e w as al so pr esen t i n both sam pl es, an d t h e m essa g e appea red t o be u n rel ate d to ei th er t h e R oy al A rm ou ri e s, or m u seu m s i n g en eral . 5 5 “ To tal l y j u st tri ed a nd fa i l e d to wa ng l e o ur wa y i n to a foo d e xh i b i ti on at Cl a re n c e Do c k i n Lee d s . Wh o k n e w th ey'd CHECK!” Fi g u re 3. Fl oor pl an of t h e R oy a l Ar m ou r i e s Mu seu m , obtai n ed f rom ra w O pen S t reet Map da ta . A l th ou gh t h i s m et h odol ogy su cceeds at col l ect i ng t w eet s g eog raph i ca l l y as soci at ed w i t h m u seu m s, i t l ar g el y f ai l s at r et u rn i ng tw eet s t h at ar e l i n ked t o m u seu m s se m ant ic all y : ju st becau s e a t w eet i s sen t f rom n ear a m u seu m , does n ot m ean i t i s abou t a m u seu m . In addi ti on , m u se u ms are of t en cl osed spac es so i t i s n ot cl ear w h et h er or n ot an accu ra te G PS posi ti on can be est abl i sh ed by th e m obi l e dev i ce bef ore tr an sm i t t i ng a m essa g e. H en ce tw eet s f r o m i n si de m u se um s m i g h t n ot be accu r at el y posi ti on ed an d appea r ou ts i de a s pat i al f i l ter . Vi si tor s com m u ni ca t ing w i th t h ei r f r i en ds a n d peopl e w h o ar e si m pl y n ear t h e m u seu m coi n ci de n t a l ly (e .g . t o u se a caf é as soci at ed w i t h th e bu i l di ng) see m t o con st i t u te t h e m aj or i t y of th es e t we ets : on l y 1 of t h e 99 tw eet s f r om t h e R oy al A r m ou ri es Mu seu m cou l d be i den t i fi ed wi th an y di r ect sem an t i c l i n k t o t h e w eapon s on di s pl ay . 6 T o ov erc om e t h ese i s su es a s em an ti c f i l t er w as u sed. T h i s m ean s sca n n i ng th e t ex t con te n t s of th e t we et s an d se l ect i ng t h ose th at con t ai n cer t ai n key w or ds. T h er e are m an y s tr at eg i es th at can be em pl oy ed t o i n cr ease th e propor ti on of rel ev an t tw eet s se l ect e d w h i l st si m u l ta n eo u s l y r edu ci n g u n w an t ed „f al se posi t i v es ‟ – tw eet s t h at ar e n ot r el ev an t to t h e t opi c u n der i n v est i g atio n . H ere, al l t w eet s con t ai n i ng on e or m or e of th e w ords „ m u seum ‟, „g al l er y ‟, „ex h i biti o n‟ an d „ex h i bit ‟ we re sel ect ed. T h i s f i l t er pro du ce d a s u bse t of 1,553 tw ee ts f rom 684 i n di vi d u a l u ser accou n t s . T h es e we re cl u st ered ar ou n d t h e u r ban cen t ers of L eeds an d B radf ord (f ig ure 5). Sel ect i ng an appr opri at e l ev el of i n v est i g atio n f or t h e ori g i n area s wa s cri t i cal : i f t h e area s w er e too s m al l , th er e w ou l d be m an y z on es con t ai n i ng n o soc i al m edi a or soc i o- dem og raphi c dat a; t oo l ar g e an d S IM w ou l d con t ai n i n su f fi c i e nt g eog raph i ca l det ai l t o be u sef u l . T h e n u m ber of z on es con t ai n i ng sem an ti ca l ly f i l t er ed m u se u m t w eet s w as ass ess ed u si n g a spat i al i n ter se ct qu er y 7 . T h e r esu l ts f or a n u m ber of adm i n i st r at i ve sc al es are pres en ted i n T abl e 1. B ased on t h ese n u m bers , an d t h e sh ape s of t h e adm i n is t ra t iv e z on es, Cen su s wa rds w er e sel ect ed as th e adm i n is tr at i ve z on e w i t h t h e m ost appr opri at e bal an ce of g eog raph i ca l det ai l an d spat i al ag g r eg ati o n . 6 T h e tw eet i n cl u ded t h e t ex t “ Jou st ti m e! ” an d a l i n k to a ph ot o. A n oth er t w eet wa s di scov er ed f r om a 20 m bu f f er , t h at w as def i n it e l y ass oci ate d wi th a m u seu m v i s i t : “ O l i v er wi th h i s n ew g u n ” an d a l i n k t o a ph ot o of a ch i l d n ex t to an ol d can n on . 7 In th i s i n sta n c e th e „s e l e ct b y l o c ati o n‟ fu ncti o n i n QG IS wa s u se d. T abl e 1. Ge og r aph i c l ev el s con s i der ed f or u se as t h e spat i al u n i ts i n t h e SI m odel . 8 L ev el N . z on es N . z on es w i th t w eet s N . T w eet s/ z on e A v . P op. OA 4000 269 3.4 300 MSO A 191 121 7.7 6300 Cen su s w ard 65 64 14.5 18500 Con st i tu en c i es 16 16 58.0 75000 L oca l A u th or ity 6 6 154.7 200000 T T W Z on e 5 5 185.6 240000 I n m os t cases i t can be ass u m ed t h at peopl e do n ot se n d tw eets f r om t h ei r h om e l ocat i on s abou t m u s eum s . A u ser ‟ s h om e l ocat i on wa s assu m ed t o be w h ere th e h i g h est n u m ber of tw ee ts occu rr ed f or a u ser accou n t. U si n g a si m pl e ag g re g at e qu ery ov er l ocat i on (g en er al ize d t o 100 m et er s) an d th e u n i qu e i den t i fi cat i o n n u m ber f or accou n t h ol der s, t h e l oca ti on s w i th t h e h i g h es t t we et i n g act i v i t y w ere i den ti fi ed . T h e di st ri but i o n of t h es e f av ori te t w eet l ocat i on s, h en cef ort h ref er red t o as „h om e l oca ti on s‟ (al th ough i t i s ackn ow l edg ed th at t h ese m ay n ot al w ay s cor res pon d w i th u ser s‟ t ru e h om e l ocat i on s) i s su bst an t i a ll y l arg er, i n corpor at i ng an ar ea r ou g h ly dou bl e th e si z e of th e „m u seu m t w eet s‟ . T h e st u dy ar ea w as det er m i n ed by t h i s l ar g er s pat i al ex t en t, spe ci f i ca ll y al l 7 L oca l A u th ori t i es i l l u s t ra ted i n Fi g u re 5. 8 S e e h e re fo r d e fi n i ti o n s an d d e sc ri pt i o ns o f UK a dm i n i s trati v e z o n e s : h ttp : // w ww. on s . g ov . uk /on s /g u i de-m e th od /g e o g ra p hy /b eg inn e r -s -g u i de/ i n dex . h t m l Fi g u re 4 . O v erv i ew of th e g eog raph i ca l di s tr i but i o n of t h e sem an ti ca l ly f i l t er ed m u s eum t w eet s (r ed dots ) an d h om e l ocat i on s (g r een t ri an gl es ). T h e sh ade of poi n t s cor res pon ds t o den si ty , i l l u s tr a t i ng h i g h den si t i es i n L eeds ( cen tr e) an d B r adf or d (t o t h e w es t) . A n al y s i s of th e m ost f requ en t tw eet ers (u se rs w hos e i ds w ere ass oci ate d wi t h m or e th an 10 m u s eu m tw eet s) l ed to t h e di scov er y of repe at ed t w eet s. R ‟ s „u n i qu e‟ f u n ct i o n w as u sed t o as cer t a i n t h e propor t i o n o f r ep eat ed t w ee ts : 27% f o r a ll t w eets, in cr ea s i ng to 31% f or th e m ost f re qu en t t we et ers . Addi t i on a l an al y si s f ou n d t h at ev en f or t w eet s w i th u n i qu e te x t con t en ts , i n m an y ca ses th e on l y di f f er ence betw een on e t w eet an d t h e n ex t w as t h e h t m l code as soci at ed w i th t h e tw ee t. In ev ery cas e t h i s wa s f ou n d t o be du e to Fou rs qu ar e, w h i ch au t om ati ca l l y sen ds g eog raph i ca l t w eet s f rom cert ai n l oca ti on s (36). 9 I t i s poss i bl e t h at som e of th es e „au t o ch eck - i n ‟ m es sag es t r i g g er ed si m pl y du e t o pr ox i mi t y t o a m u seu m ra th er th an g en u i ne i n t ere st an d i t i s cl ear th at th e se t we ets are n ot of th e sa m e v al u e as t h ose w h o del i ber at el y com posed m ess ag es abou t m u seu m s an d cou l d pot en ti all y be m i sl ead i ng . T h e f ou r squ ar e t w eet s w er e i den ti f i ed an d r em ov ed l eav i ng 928 u n i qu e t we et s rem ai n i n g . T h ese we re t h e dat a poi n t s u sed to i n f or m t h e SIM. A S pat i al Interact i on mode l of m useum vi sit s I n th e absen ce of of f i c i a l v i s i t data f rom m u s eum s i n t h e st u dy ar ea, a SIM can n ot be pr eci sel y con st rai n ed i n ter m s of f l ow s by dest i n at i o n . D at a on th e f l ows em an ati ng f r om eac h ori g in i s scar cer st i l l . A n ot h er i ss u e to con si der i n t h i s su b- n at i on a l cas e st u dy i s t h at n ot al l m u se u m dest i n at i o n s are con si dered: on l y t h ose i n L eeds an d B radf o r d L oca l A u th or iti es (s ee f i g u r e 5 ) – al t h ou gh L eeds i s som ew h at of a reg i on a l cen t er, m an y r esi den ts , espec i al ly t h os e l ocat ed f ar f rom l ocal m u seum s wi l l tr av el f u rt h er af i el d. 10 For t h ese reas on s an u n con st rai n ed SI M w as u sed (11) . B u i l d ing on equ at i on 1, t h e f l ows (T ij ) w er e est i m at ed as f ol l ow s : (4) 9 Th i s c a pab i l i ty , an d h ow to di s a b l e i t, i s fu rth e r d e sc ri bed o nl i ne: h ttp : // a bo ut fou rsq uare .c om /d o-y o ursel f-a -fa vo r-and - st op -s e n di ng -every -c hec k i n- to -twi tter- and -fa ce bo o k / Wh er e In ci i s t h e i n com e- adj u st ed dem an d f or m u seu m tr i ps per u n i t popu l at i on (P) i n eac h z on e an d β t h e di st an ce- decay par am et er i n tr odu ced ear l i er . W j i s t h e „ att r act i v e n ess ‟ of m u seu m j, cal cu l ate d as a c om pos i t e of f ac tor s. Est i m at i n g mus eum att racti venes s I t i s cl ear th at a w i de ran g e of f act or s i n f lue n ce h ow „at t rac ti v e ‟ m u se u m s ar e. Som e peopl e w i l l tr av el to m u se um s becau se th ey h av e a speci al i st i n t ere st i n t h e obj ect s on di spl ay ; ot h ers w i l l seek ou t m u seu m s base d on f am i ly con si der at i o n s ; w h et h er t h er e ar e f ac i l i t i es f or ch i l dren or cooked f ood, f or ex am pl e. Q u an ti tativ e l y , at tr act i v en es s can be se en as a coef f i c i e n t (W) t h at i s a f u n cti o n of v ar i ou s c on tr i bu t ing f ac tor s: (5) w h ere X i s a v ect or of f act ors rel at ed t o m u se um at tr act i v en es s an d a1, a2 et c. are coef f i c i e n t s re pres en ti ng h ow i m por ta nt each i s con si dere d ( 13). T h i s al l ow s t h e at t ract i v e n ess of each m u se u m t o be det erm i n ed by an y n u m ber of f act ors. In ou r m odel w e u se on l y tw o f act or s: f l oor ar ea an d n u m ber of m en ti on s i n t h e m edi a based on se arch es i n G oog l e New s. T h e f orm er i s an i n di catio n of capac i t y (an d cou l d be f u rt h er r ef i n ed, f or ex am pl e by i n cl u d i ng n u m ber of f l oor s) ; t h e l at ter i s an i n di cat i o n of h ow w el l - kn ow n th e m u seu m i s. T h ere i s g r eat v ari abi li t y i n t h ese prox i es of at t rac ti v en es s, as i l lu s t ra ted i n T abl e 2. It i s cl ear th at t h e N at i on al Medi a Mu se u m an d th e R oy al A rm ou ri es ar e by f ar th e m ost w e ll - kn ow n . We u sed a m odi f i ed v ers i on of equ ati on (5) t o det erm i n e at t ract i v e n es s : (6) w h ere FA i s f l oor area an d MM i s n u m ber of m en t i on s i n t h e m edi a, n orm al i z ed t o on e. T abl e 2 : Mu se u m ch ara ct eri st i cs an d p rox i es of at tr act i v e n ess. M u s eum T w e e t cou nt M ea n h o m e - m us e um d i s t . (k m ) A v erage t w e e t- m us e um d i s t . (k m ) M u s eum flo or p l a n (m 2 ) N e w s M ent i on s A b b e y H o u s e M u s eu m 8 2 .9 132 1072 2 A rm l ey M i l l s 55 3 .5 194 2734 2 B r a d fo rd In d u s t ri a l M u s eu m 11 5 .6 110 1382 1 C artwr i g h t Hal l 2 8 .5 95 1519 4 H e n ry M o o re Ins titu t e 25 6 .6 86 562 5 L e ed s A rt G a l l e ry 93 5 .5 115 1322 8 L e ed s C i ty M u s e u m 102 5 .2 130 1731 7 N a t i o n al M ed ia M u s eu m 288 8 .5 131 3211 252 R o y a l A r m o u ri es 154 6 .4 134 5180 36 Th a ckr ay M e d i c al M u s eu m 18 1 3 .7 136 1790 5 Est i m at i n g dem and D em an d f or m u se u m v i si ts can be s een as a com posi te of t h e tot al popu l at i o n of an area an d th e i n h abi ta nt s‟ prope n si ty to v i si t m u se u m s. I n t h e m odel a w ei g h t ing f act or (I n c ) w as u sed t o w ei g h t t h e popu l at i on of each w ard: (7) Wh er e A rt s i s th e propor t i on of i n di vid u a ls w h o f r equ en t „f i n e ar ts ‟ est abl i shm e n t s an d E a i s a pr ox y of av er ag e ear n i n g s , i n eac h w ard ar ea . T h ese dat a ar e g en e rat ed f rom a su bst an t i a l l i f est yle su rv ey of t h e U K popu l ati on (37) . U n su rpr i s ingl y , A rt s an d E a v ari abl es ar e cor rel at ed, al bei t w eakl y (r = 0.20) – h i g h er ear n ers m ore f re qu en t l y v i s i t f i n e ar ts at t rac ti on s. A s w i th W, I n c w as di v i ded by i ts m ea n to set t h e m ean equ al t o on e, al l ow ing di rect com par i s on s be tw een th e m odel r u n s wh i ch do an d do n ot accou n t f or at tr act i v en e ss an d dem an d. Inf err i ng fl ows fr om Tw i t t er dat a G eog ra ph i c tw eet s are di s cret e ev en t s t h at occu r i n con ti n u o u s space an d t i m e w i th a h i g h deg re e of r an dom v ari abil i t y . Fl ow m odel s , by con tr as t, u se m ath em at i cs t o des cri be sm ooth di st ri but i o ns ov er spa ce an d t i m e. E v en wh en f l ow m odel s ar e st och as ti c, t h e u n der l ying probabi l i t i es g en eral l y re pres en t sm ooth di st ri but i o n s (38) . I n t h i s con t ex t i t i s cl ea r t h at spat i al an d t em por al ag g r eg at i o n i s n eede d t o l i n k T w eet s w i th f l ow dat a. T h e t em poral ag g reg ati o n u sed i n t h i s ex am pl e i s si m pl e : t h e su m of al l t w eet s ov er th e 445 day s of col l ect i o n . T o coi n ci de w i t h th e SIM , tw eet s w er e al so ag g reg at ed s pati al l y , al l oca te d to a w a rd ass oci at ed w i t h t h e u ser ‟s m os t f re qu en t t w eet i n g l ocat i on an d m u s eu m s, on e f or each tw ee t sen t . T h i s pr oces s of spat i al ag g reg ati o n i s a key st ep tow ar ds i n f er r i ng f l ow s f r om v ol u n t eer ed g eog raph i ca l dat a, an d i s sh ow n i n f i g u re 5. Fi g u re 5. Fl ow m aps of i n f er red m u s eu m v i si t s f r om r aw t w eet s (a bov e) an d f r om spa ti al ly ag g reg at ed t w eet h om e l ocat i on s (bel ow ). Compar i ng mode l out put wi t h t w eet s T h e proces s of spat i al ag g r eg at i o n pr esen t ed i n Fi gu re 5 i s n ot on l y u sef u l f or v i s u alizing t h e den si ty of f l ow s; i t can al s o be u sed to g en er at e an ori g i n- destin a t i o n f l ow m at r i x w i t h th e sam e di m en s i o n s as T i n equ at i on 4 f or di re ct com par i son w i t h th e SIM , w h i ch w e sh al l r ef er to a s T ‟ . T h e pse u do code u s ed t o g en er at e T ‟ i s pres en te d bel ow : T ’ < - matrix (0 , nrow=nrow(S), ncol=ncol(S)) # initial conditions for i in 1: n.tweets { T ’[home[i], c.museum[i]] < - T ’[home[i], c.museum[i]] + 1 } Wh er e h om e[ i ] repr esen t s t h e h om e w ard of th e pers on assoc i at ed w i th t w eet i an d c.m u seu m [ i ] i s th e i n dex of th e m u se u m cl os est to t h e g eog ra ph i ca l coordi n at es of t w eet i re spect i v e l y . T h u s T ‟ i s a spar se m atr i x con t ai ning posi t i v e i n t eg ers on l y f or t h e or i g i n - des t i n atio n f l ow s r epres en t ed i n f i g u re x an d th e su m of T ‟ i s equ al t o th e tot al n u m ber of tw eet s (756 i n t h i s ca se) . From t h i s poi n t com par i son of T an d T ‟ can be ach i ev ed by con cat en at i ng th e m at ri ces i n t o v ect ors an d u si n g st an dard m ea su res of f i t , su ch as Pears on ‟s coef f i c i e n t of cor rel ati on (r ) an d root m ean squ ar ed (r m s) . R e s u l ts : cor r e l ati on s an d in con s is te n ci e s b e tw e e n T w i t te r d ata an d m od e l ou tp u t Basel i ne m odel T h e basel i n e s cen ari o con si st ed of t h e si m pl es t i m pl em e n t a t i o n of equ at i on 4, w i t h th e va ri abl es In c an d W set to 1 f or al l w ar ds an d m u se u m s re spect i v e l y . T h u s, t h e on l y v ari abl e to opt i m i z e wa s β, w h i ch w as i n i t i a l l y set t o 0.3 , re su l t i ng i n a pos i t i v e cor rel ati on of 0.31 bet w een t h e 2685 v al u es of T an d T ‟ . It er at i n g th r ou g h 200 m odel - t w eet obser v at i on com pari son s (s te p si z e = 0.01) i t wa s f ou n d t h at m odel f i t wa s opt i m al w i t h a β v al u e of 0.95, w i t h th e corr el at i on peaki n g w i t h an r v al u e of 0.39 . Inclu di n g mus eum at t rac t i veness T h e n ex t m odel t est ed added t h e W v ari abl e – s et by equ at i on 6 – to t h e m odel , to m ake l arg er an d m ore f r equ en t l y m en ti o n ed m u se u m s m ore at t rac ti v e th an sm al l m u se u m s t h at f ew peopl e h ad h ear d of . T h e i m pact of th i s ch an g e w as dr am at i c, w i t h r v al u es peaki n g at 0.60. T h e opti m al di st an ce decay f u n ct i o n i n th i s m odel s peci f i cat io n w as f ou n d t o be st eepe r (β = 1.32 ). Inclu di n g dem and an d at t racti venes s I n t h e fi na l m ode l te st t h e added r e f in e m e n t o f v a r i ab l e de m a n d f ro m t h e o r igi ns w as s e t , by al t eri ng I n c v al u es accor di n g t o equ at i on 7. T h e per f or m an ce of th i s m odel spec i f i catio n w as f ou n d t o be i n ter m ed i at e com pare d wi th t h e oth er tw o , wi th a m ax i m um r v al u e of 0.55 . T h es e res u l t s ar e di spl ay ed i n f i g u re 6 bel ow. Fi g u re 6. C orr el at i on bet we en th e spat i al i n t er act i o n m odel ( S) an d f l ow s i n f er red f r om t w eet s (S‟ ) ag ai n st β v al u es, f or th r ee di f f er e n t m odel s peci f i cat io n s. M odel enhan cem ent s T h er e ar e m an y re f i n e m e nt s t h at can be m ade to t h e bas i c m odel s s peci f i ed abov e. A m ajor ch an g e w ou l d be t o u se a con st ra i n ed spa ti al i n t er act i o n m odel , w h ere by t h e f l ow s f rom each ori g i n are set (s ee f i g u r e 7). I t w as f ou n d t h at t h e m odel f i t decl i n ed g rea tl y i n th i s sce n ari o, h owe v er, du e t o f or ced f l ow s t o di s ta n t m u se u m s f r om w ar ds i n t h e peri ph ery of t h e st u dy r eg i on . T h e decr eas e i n m odel f i t i n th e con st r ai n ed m odel f i t w as ex pect ed f or t w o re ason s : T h e st u dy ar ea i s bei n g t rea te d as an en capsu l at ed sy st em w h i ch i s art i fi c i a l ly i m posi ng a bou n dary t h at does n ot ex i st i n t h e re al wor l d on th e data . T h i s h as t h e ef f ect of i n fl at i ng f l ow s ori g i n a t ing n ear th e edg e of t h e st u dy area w h ere al te rn at i ve att r act i on s a re bei n g ex cl u ded becau s e th ey f al l on th e w ron g si de of t h e bou n dar y . T h i s ef f ect i s m ore com m onl y kn ow n as t h e bou n dar y ef f ect . It is al so l i kel y t h at peopl e l i ving i n r em ot e ar eas si m pl y v i s i t m u s eu m s l es s f r equ en t l y du e t o acces si bil i t y . Fi g u re 7. C on st rai n ed an d u n con st r a i n ed v ers i on of th e m odel wi t h W set . T h e g re en t ri an gl es are 20 ra n dom l y sel ect ed or i g i n s . Map pr oje ct i on: O SGB 1936. A n u m ber of f u rt h er r ef i n e m e nt s cou l d be m ade to t h e est i m ati o n of dem an d an d at t ract i v e n ess , by f u rt h er adj u st m en t s to equ at i on s 6 a n d 7. Mor e ex peri m ent s h av e been u n dert aken , bu t i t i s con si dered th at t h e data i s i n su f f i c i e n t l y l ar g e to r el i abl y cal i br ate m u l t i p l e par am ete rs w i t h ou t u n desi rab l e „ov er f i t t ing‟ ef f ect s, w h ereby a m odel i s es sen t i al l y speci f i ed t o f i t n oi se w i t h in th e dat a. D u e t o th e rel at i v e l y sm al l sam pl e si z e of m u se um t we ets u se d i n t h i s s tu dy th e m odel w as des i g n ed si m pl i st ic a l l y an d i t w as deem ed u n n ecess ary an d poten t i ally u n su pport abl e to f u rt h er com pl i c at e t h e m odel w i th addi ti on a l para m et er s. In st ead, th e res u l t s f orm t h e basi s of a di scu ss i o n of th e pote n t i al f or V G I f r om soci al m edi a t o i n f or m spat i al i n t erac ti o n m odel s i n g en eral ter m s. B ase d on t h i s di sc u ss i o n , ou r m et h ods ca n be s een as a st eppi n g st on e tow ards cal i brat i ng m ore com pl ex m odel s based on l arg er V G I dat aset s an d m ore pres si n g rea l- w orl d appl i ca ti o n s . D i s cu s s i on an d con cl u s ion I n th i s pape r w e h av e u sed g eo - l ocat ed tw eet s to c al i br ate a spat i al i n t erac t i o n m odel r epres en ti ng v i si tor f l ow s t o m u seu m s f r om res i den t i a l ar eas su rr ou n d ing L ee ds an d B r adf or d i n th e U K . In th e proce ss w e h av e h i g hl ight ed t h e pot en ti al f or VG I as an i n pu t i n t o qu an ti t at iv e g eo g raph i c a l re sear ch i n g en er al , an d t h e u s e of T w i tt er dat a t o cal i brat e SI Ms i n par ti cu l a r . Met h ods f or f i l t er i ng an d ag g re g at i ng th e g eo - l oca ted t w eet s h av e been dev el oped, a l l ow ing th e SIM t o be cal i brat ed by i n f or ma t i o n m ade pu bl i cl y av ai l ab l e on t h e i n t ern et . A l t h ou gh th e scope of th e cas e st u dy i s l i m i t ed g eog raph i ca l ly to on e st u dy ar ea an d sem an t i ca l l y t o m u seum s, t h e st u dy dem on st rat es t h e wi der poten t i al of V G I f rom soci al m edi a i n g eog ra ph i ca l re sear ch an d pr ov i des a f ou n dat i o n f or di sc u s si o n of V G I i n g eo- sci en ce appl i cat i o n s ov er al l . D at a l i mi t at i on s I n ag ree m en t wi t h Goodc h i l d (3) , we con cl u de t h at f r ee, g eog ra ph i c an d sem an ti ca l ly r i ch dat ase ts deri v ed f r om t h e h ar v est ing of soci al m edi a s i t es en m asse sh ou l d con t i n u e t o be of g rea t an d g r ow i n g i n t er est t o al l spat i al an al y st s . Da ta qu al i t y rem ai n s a ser i ou s con cer n f or al l su ch VGI , h ow ev er (2) . T h e sh eer di v ers i t y an d spor adi c n at u re of th e dat a pose s n ew c h a l le ng es t o r esear c h ers acc u st o med t o re l at ive ly c l ea n o ff i c i a l da t a se t s . A s em ph as i z ed t h r ou g h o ut, th es e ch al l eng es sh ou l d n ot be ov erl ooked: t h ey m u s t be ackn ow l edg ed at t h e ou ts et an d t ackl ed w i t h care. Mor e spe ci f i ca ll y , t h e m ai n l i m i t a t i o n s of t h e dat a u sed i n t h i s s tu dy i n cl u de : l i m i ted dat a av ai l ab il i t y t h e su b n at i on al n at u r e of th e st u dy area a l i m i te d set of m u seu m s bei n g con si der ed, an d u n cer tai n t y abou t th e tr av el beh av i ou r of th os e peopl e l i vi ng re m ot el y t o m u s eu m at tr act i on s. I m pr ov ed dat a h arv es t ing an d r etr ospe cti v e dat a col l ec ti o n cou l d h el p ov ercom e th e f i rst t w o i ssu es ; u se of of f i c i a l ly re g i st er ed m u se u ms an d v i si t or dat a cou l d t ackl e t h e th i rd an d f ou rt h poi n t s. T h e broade r probl em of w i de sem an t i c di v er si t y can be par ti al ly t ackl ed by i n t el lig e n t f i l t er i ng , u s i n g sear ch t erm s t h at capt u r e on l y m essa g es th at can be con f i dent l y at tr i bu t ed t o t h e t opi c of i n t er est . N ev ert h el ess , t h ere w i l l i n ev it ab l y st i l l be v ar i at i o n s i n qu al i t y an d i ssu es of sel f - sel ect i o n ev en af t er th e m os t st ri n g e n t f i l t e r ing st r ate g i es. On e w ay of ta ckl i ng t h i s i ss u e w ou l d be to a ssi g n a con t i n u o u s we i g h t v ari abl e to ev er y t we et cor res pon di ng to i t s re l ev an ce t o th e res ear ch probl em , al t h ou gh t h e m et h od of ass i g n ing h i g h an d l ow w ei g h t s w ou l d add f u rt h er com pl exi t y an d su bj ect i vit y t o th e an al y si s . T h i s poi n t i s r el at ed t o a m ore f u n dam e nt a l probl em w i t h soc i al m edi a dat a: t h e con t ex t- depen den ce. T w eet s sen t to an d f rom i n di vi d u a ls con ta i n m an y s u bt l eti es th at ar e u se f u l t o u s ers wh o can decode th em . “ Y et , take n ou t of con t ex t, da ta l ose m ean i ng an d v al u e” (39) . In th e proc ess of h arv es ti ng an d su bse qu en t sa m pl ing m u ch of t h i s con tex t i s l os t: cu rr en t ly th er e i s n o w ay to con si der th e w i der con t ex t, i m pli c it i n eac h t w eet , ov er th ou san ds an d i n deed m i l l io n s of su ch dat a poi n t s. Fu t u re dev el opment of VG I der i ved f rom soci al m edi a T h e rapi d pen et rat i on of m obi l e ph on es w orl dw i de i s bei n g cl osel y f ol l ow ed by u pt ake of sm ar t ph on es, m an y of w h i ch con t ai n G PS r ecei v ers an d t h e capac i t y to i n t er act wi th pu bl i c soc i al m edi a w ebsi te s su ch as T w i tt er , Facebook. T h ere f ore th e pot en ti al f or g row t h of dat ase ts su ch as th e g eo- ta g g ed t w eet s pr esen t ed i n t h i s pa per i s tr u l y en orm ou s, w i t h equ al l y h u g e pot en t i al f or res ear ch ers (40) . T o som e ex t en t t h e af orem en t i o n ed dat a l i mi ta t i o ns cou l d be part l y m i ti gate d by sh eer v ol u m e, i n cre asi ng t h e si g n al - to - n o i se rat i o ( 41). H ow ev er, con ti n ued n ear- ex pon en t i a l f u t u re g r ow th i n pu bl i cl y av ai l ab l e g eo- ta g g ed soc i al m edi a i s by n o m ean s ce rt ai n : i ss u es of dat a pr i v acy an d ow n er sh i p cou l d h am per th e av ai l ab i li t y of th es e dat aset s to r esea rch er s w orki n g i n t h e pu bl i c dom ai n . T h er e i s a dan g er t h at com pan i es su ch as T w i tt er re st ri ct acces s to t h e dat a on l y t o pri v at e com pan i es w i th t h e f i n an c i a l res ou rces to pa y f or th e acce ss. T h i s i s to som e deg re e bei n g m i t ig a ted by i n i t ia t i ves t o m ake pu bl i cl y av ai l ab l e soci al m edi a m es sag es av ai l ab l e th r ou g h pu bl i cl y own ed cen t ral i z ed r eposi tor i es su ch as t h e L i bra ry of Con g re ss, w h i ch sh ou l d al l ow f or ret r ospect i v e s ear ch es, r at h er th an res ear ch ers ‟ cu rr en t rel i an ce on r eal - t i m e h arv es ti ng t h rou g h A PIs an d ret r ospect i v e sc rapi n g of dat a f r om t h e i n t ern et (42) . A n oth er i ssu e i s th e tr an si t ory n at u re of soci al m edi a we bsi te s: i t h as been h y poth es iz ed t h at popu l ar soc i al m edi a si te s su ch as T wi t te r an d Facebook con t ai n t h e seeds o f th ei r ow n dem i s e, so i t wou l d be a m i s ta ke to as su m e u n cr i t i ca ll y t h at a si n g l e ser v i ce can be r el i ed u pon f or soci al V G I i n to th e l on g- te r m . In addi ti on i n cr eas ed pu bl i c aw ar en ess of di g i ta l dat a ar ch i vi ng an d an al y s i s i n th e w ake of l eaked N at i on al Secu ri ty A g en cy (NS A ) docu m en ts cou l d l ead t o a cu rt ai l me n t or al t erat i o n i n t h e sh ari n g of pers on al i n f or m atio n su ch as l ocat i on . Al t ern ativ e l y , i t cou l d be arg u ed t h at m aki n g cer ta i n com m u ni ca t i o n s ex pl i c i t l y pu bl i c, f or soci al ben ef i t, c ou l d becom e i n cr eas ingl y com m on as a cou n t erba l an c e to t h e perce i v ed con cen t ra ti o n of di g i ta l i n t el l ig e nce g at h er i ng an d pr ocess i n g capa bi l i t i e s by a f ew cl an dest in e org an i z atio n s. Fu t u re r esear ch pot ent i al D espi te t h e data l i m i ta t i o ns, et h i cal con cer n s an d u n kn ow n l on g evi t y of soci al V GI , i t se em s l i kel y th at t h e si z e an d ri ch n ess of av ai l ab l e data set s w i l l con t i n u e to g row . In par al l el wi t h th i s, com pu ti ng pow er w i l l con ti n u e to i m pr ov e an d com pu te r pr og r am s w i l l con t i n u e to dev el op t ow ards g r eat er f u n ct i o n a l ity an d u ser f r i en d li n es s. T h i s m e an s VGI f r om s oci al m edi a wi l l becom e an i n cr easi ngl y at tr act i v e al t ern at i v e t o of f i ci a l dat ase ts f or g eog ra ph i c a l probl em s t h at are pr ese n t ed i n t h i s paper, w h ere data l i m i t a t i o n s rem ai n a m ajor con st ra i n t . T h er e i s cl ear l y g re at pot en t i al f or f u rt h er r esear ch u s i n g t h i s dat as ou rce i n m an y area s, i n cl u d i ng th e f ol l owing: G eog raph i ca l l y ex t en si v e ev en t s su ch as r i ot s (43) an d epi dem i cs (44). T h e an al y si s of sh i f t i ng at ti t u des an d beh av i or s i n rel ati on to h eal th or en v i r onm e nt a l dri v er s (45) . T h e u se of soci al - m edi a VGI t o i n f or m an d cal i bra te a - pri ori m odel s of spat i al beh av i or su ch as th at pr esen t ed i n t h i s pa per. B eca u s e th e ex pl osi on of V GI i s a re cen t ph en om en o n t h at ch al l eng es es tabl i sh ed h abi ts of doi n g re sear ch , th er e i s h u g e u n t apped pot en ti al i n each of th es e ar eas an d m or e. We u rg e r esea rch er s t o t h i n k car ef u l ly abou t th e res earc h probl em s th at t h ey are i n te res te d i n bef ore ev al u at i ng w h et h er or n ot g eot ag g ed soci al m edi a i s an appr opri at e i n pu t da ta s ou rce. In m an y cas es i t wi l l n ot be: i t i s i m por tant th at re sear ch er s do n ot si m ply f ol l ow th e n ew est an d l ar g est dat ase ts j u st becau se th ey ar e av ai l ab l e an d (pot en ti al l y ) f r ee at sou r ce. It i s h oped th at th i s paper w i l l l ea d to f u rt h er di scu ss i o n of th e rel ati v e m er i ts of T w i t t er an d ot h er v ol u n te e red soci al m edi a i n f or m atio n f or i n f or m ing g eog raph i ca l res ear ch . E t h i ca l con si der ati o n s sh ou l d al so g u i de t h e res ear ch : i f t h e i n f or m at i o n i s pr ov i ded by t h e pu bl i c f or f ree, su re l y t h e ben ef i t s t h at accr u e sh ou l d be f or pu bl i c be n ef i t . R e fe r e n ce s 1. E l w ood S, Goodc h i l d MF, Su i D Z . R es ear ch i ng Vol u n t eer ed G eog ra ph i c I n f orm at i o n: S pat i al D at a, G eog raph i c R es ear ch , an d N ew Soci al P ra cti ce. A n n al s of th e As soci at i o n of A m er i can G eog ra ph er s. 2012; 102(3) : 571 - 90. 2. Fl a na gin A J, Me t zg er MJ . Th e cr ed i b i li t y o f v o l unte er ed g eo gra p hi c i nfor m at i o n . G eoJou r n al . 2008; 72(3- 4): 137- 48. 3. G oodch i l d MF. C i ti z en s as sen s ors : t h e wor l d of v ol u n t ee red g eog r aph y . G eoJou r n al . 2007; 69(4) : 211- 21. 4. Cl ar ke G . C h an g i ng m et h ods of l ocat i on pl an n i ng f or r eta i l com pan i es . G eoJou r n al . 1998(1977) : 289- 98. 5. N akay a T , Foth er i n gh a m aS, H an aoka K , C l arke G , B al l as D , Y an o K . Com bi n i ng m i cr osi m ul a t i o n an d spat i al i n ter act i o n m odel s f or ret ai l l ocat i on an al y si s. Jou rn al of G eog raph i ca l Sy st em s. 2007; 9(4) : 345 - 69. 6. Wi l s on A G . L an d- u se/ tr an sp ort i n t era ct i o n m odel s: P as t an d f u t u re. J ou rn al of T r an spor t E con om i cs an d Pol i cy . 1998;32( 1) : 3 - 26. 7. R ae A . From spat i al i n t erac ti o n dat a to s pati al i n te racti o n i n f orm a t i o n ? G eov i su a l i sat io n an d spa ti al st ru ct u res of m i g r at i o n f r om t h e 2001 U K cen su s . 2009. p. 161- 78. 8. H e J, Pool er J. Model i n g C h i n a's pr ov i n ce- to - prov i nce m i g rat i o n f l ow s u si n g spa ti al i n t era cti o n m odel w i th addi ti on a l v ari abl es . G eog ra ph y R es earc h Foru m . 2003: 30- 55. 9. Z u o C , B i rki n M, C l ar ke G , McE v oy F, B l oodw ort h A . Model l ing th e t ran s port at i o n of pri m ary ag g reg at es i n E n g l an d an d Wal es : E x pl or ing i n i t i at i v es t o redu ce CO 2 em i ssi o ns. L an d U se P ol i cy . 2013; 34: 112 - 24. 10. W il so n AG . A f a m il y o f spat i a l i nte r act i o n m ode l s, a n d as soc i a t ed de v e l op m e nts . E n v i ro nm e nt an d Pl an n i ng . 1971; 3 (Ja n u ar y ) : 1- 32. 11. Fot h er i n gh a m AS. Spat i al s tr u ct u r e an d di st an ce- decay par am ete rs. An n al s of t h e As soci at i on of A m er i ca n G eog raph er s. 1981; 71( 3): 425 - 36. 12. O pen sh aw S. N eu ral n et wor k, g en et i c, an d f u z z y l og i c m odel s of spat i al i n t er act i o n . E n v i r o nm e n t an d Pl an n i ng A . 1998; 30(1970) : 1857- 72. 13. Wi l s on A G . C om pl ex spat i al s y st em s : th e m odel l i ng f ou n dat i o n s of u rba n an d r eg i on al an al y s i s : Pear son E du cat i on; 2000. 14. O pen sh aw S. A n em pi ri ca l st u dy of som e spa ti al i n t era ct i o n m odel s. 1976; 8(1) : 23- 41. 15. Di pl ock G , Open sh aw S. Us i n g si m pl e g en et i c al g or i t hm s t o cal i brat e spa ti al i n t er act i o n m odel s . Ge og r aph i ca l A n al y s i s . 1996; 28(3) . 16. D en n et t A . E st i m at i ng f l ow s betw een g eog raph i ca l l oca ti on s : ' g e t m e st ar t ed i n 'spa ti a l i n t er act i o n m odel l ing . U C L Wor ki n g Pa per s Seri es. 2012;44( 0): 0- 24. 17. Joh an ss on B r, K l ae sson J, Ol s son M. Com m u ter s' n on- l i n ea r res pon se t o ti m e di st an ces. J ou rn al of G eog raph i ca l Sy st em s . 2003;5( 3): 315 - 29. 18. O ‟Kel l y M, N i edz i el sk i M, G l eeson J . Spat i al i n t er act i o n m odel s f r om I ri sh com m u t ing dat a: v ar i at i o n s i n t ri p l en g t h by occu pat i on an d g en der. J ou r n al of G eog ra ph i ca l Sy st em s. 2012; 14(4) : 357 - 87. 19. Wu B , C ai L a. Spat i al Model i ng . An n al s of T ou ri s m R esea rch . 2006; 33(1) : 179 - 98. 20. H al l CM. Space t i m e acces si bi li t y an d t h e T A L C: T h e rol e of g eog r aph i e s of spa ti al i n t era cti o n an d m obi li t y i n con t ri bu t i ng to an i m prov ed u n der st an d i ng of t ou ri sm . In : B u tl er R , edi t or.: C h an n el V i ew Pu bl i cati o n s ; 2006. p. 83 - 100. 21. Papi n s ki D , Scott D M. R ou te ch oi ce ef f i c i e n c y: an i n v est i g at i o n of h om e - to - w ork t ri ps u si n g GPS dat a. E n v i r o nm e nt an d Pl an n i ng A . 2013; 45(2) : 263 - 75. 22. V i n e D , B u y s L , A i r d R . T h e u se of am en i t i es i n h i g h den si t y n ei g h bo u r h ood s by ol der u rban A u st ral i a n re si den ts . L an dsca pe an d U rba n Pl an n i ng . 2012; 107(2) : 159- 71. 23. Fot i s J, B u h ali s D, R oss i des N . Soci al m edi a i m pac t on h ol i day t rav el pl an n i ng: T h e case of th e R u ss i an an d t h e FS U m ar ket s. I n ter n at i o n a l Jou rn al of O n l i n e Mar ket i n g (I JO M) . 2011; 1(4) : 1- 19. 24. H v ass K a, Mu n ar aM. T h e t akeof f of soci al m edi a i n t ou ri sm . Jou rn al of V acat i on Mar keti n g . 2012; 18(2) : 93- 103. 25. B on sal l P, edi t or. T ra v el l e r beh av i or: D eci si on - m ak i ng i n an u n pre di ct ab l e w orl d. 2004. 26. G oh YM , L ov e PE D . Met h odol og i ca l appl i cat i o n of sy st em dy n am i cs f or ev al u at i ng t raf f i c saf ety pol i cy . Saf et y Sci en ce. 2012 ; 50(7) : 1594 - 605. 27. Sav ag e M, B u r row s R . T h e com i n g cr i s i s of em pi ri ca l soci ol og y . Soci ol og y . 2007; 41(5) : 885- 99. 28. B i rki n M, Ma l l eson N . T h e Spati al A n al y si s of Sh ort- te rm Popu l at i o n Mov em en t s w i t h S oci al Medi a D ata . 29. Ch en Y - Y , Su el T , Ma rkow et z A , edi t ors . E f f i c i e nt qu ery proces si n g i n g eog raph i c w eb se arc h en g i n es . 2006; N ew Y or k, N Y , USA : A C M. 30. R u ss el l MA . Mi n i ng th e Soci al Web: An al yzi ng D at a f rom Facebook, T w i tt er , L i n kedIn , an d Oth er Soci al Medi a S i tes : O 'R ei l l y Medi a; 2011. 31. Packer HS, Sam an g ooei S, Ha re JS, G i bbi n s N , L ew i s PH , edi tor s. E v en t det ect i on u si n g T w i t t er a n d st ru ct u red s em an ti c qu ery ex pan si on . 2012; N ew Y ork, N Y , U SA: A C M. 32. Sakag u ch i T , A kah o Y , T akag i T , Sh i n t ani T , edi t or s. R ecom m en dat i o n s i n T w i tt er u si n g con cept u al f u z zy se ts . 2010. 33. V er m a S, V i ew eg S , Corv ey WJ, Pal en L , Ma rt i n JH , P al m er M, et al ., edi tor s. N at u ral L an g u ag e P roc ess i n g to t h e R es cu e? E x t rac ti ng" Si tu at i o n a l A w are n es s" T w eet s D u ri ng Mas s E m er g en c y . 2011. 34. A bel F , Cel i k I, H ou ben G- J, Si eh n del P. Le v er ag i ng th e se m an t i cs of tw ee ts f or adapt i v e f acet ed sear ch on t wi tt er . Spri n g er ; 2011. p. 1 - 17. 35. R oodh ou se S. Com pan y Prof i l e : A C h al l eng e t o C u l t u ra l Sect or M an ag em e n t Con v en t i o n s : T h e R oy al Ar m ou r i es Mu se u m . In t er n at i o n a l Jou r n al of A rt s Man ag em ent. 1999: 82 - 9. 36. L i n dqvi s t J, C ra n s h aw J , Wi es e J, H on g J, Z i m m er ma n J, edi t ors . I'm th e m ay or of m y h ou se: ex am i ni ng w h y peopl e u s e f ou rs qu are - a s oci al - dr i v e n l oca ti on sh ar i n g appl i cat i o n . 2011. 37. T h om pson C , Sti l l w e l l J, C l arke M, B ra dbrook C . U n ders tan di ng an d Va l i dat ing A cx i om ‟ s R esear ch O pi n i on P ol l D at a. L eeds Wor ki n g Paper s: Un i v ers i t y of L ee ds; 2010. 38. Si m i ni F, G on z al ez MC, Mar i t an A , B ar abasi A L . A u n i v er sa l m odel f or m obi l i ty an d m i g ra t i o n patt er n s. N at u r e. 2012; 484(7392) : 96 - 100. 39. B oy d D , C raw f ord K . C ri t i ca l Q u es ti on s f or B i g D at a. I n f orm at io n , Com m u ni ca t i o n & Soci ety . 2012; 15(5) : 662 - 79. 40. Cram pt on JW , Gr ah am M, Poort h u i s A , Sh el ton T , Steph en s M, Wi l s on MW , et al . B ey on d th e g eot ag : s i t u at ing 'bi g dat a'an d l ev era ging th e pot en ti al of th e g eow eb. SSR N E l ec tr oni c Jou r n al . 2013: 1 - 29. 41. H akl ay MM, B asi ou ka S, A n ton i o u V , A th er A . H ow m an y v ol u n t eer s does i t t ake to m ap an are a w el l ? T h e v al i dity of L i n u s‟ l aw to v ol u n te ere d g eog raph i c i n f or m atio n . Cart og raph i c Jou r n al , T h e. 2010;47( 4): 315 - 22. 42. H i tch coc k S . Col l ect i ng an d ar ch i vi ng t w eet s: a D ata Pool cas e st u dy . JI SC D ata Pool . 2013(M ay ): 1- 12. 43. Proc ter R , V i s F, Vos s A . R eadi n g th e ri ot s on T w i tt er : m eth odol ogi ca l i n n ov at i o n f or th e an al y si s of bi g dat a. In t ern at i o n a l Jou r n al of Soci al R esea rch Met h odol og y . 2013;16: 197 - 214. 44. Ch ew C , E y sen bac h G . Pan dem i cs i n th e ag e of T w i t t er: con te n t an al y si s of T w eet s du r i n g t h e 2009 H 1N 1 ou t break. Pl os O n e. 2010;5( 11): e14118 - e. 45. K am el B ou l os MN, San f i li ppo AP, Corl ey C D , Wh eel er S . Soci al Web m i ni ng an d ex pl oi ta t i o n f or se ri ou s appl i ca ti o n s : T ech n osoci a l Predi cti v e A n al y t i cs an d r el at ed t ech n ol ogi es f or pu bl i c h ea l t h , en v i r onme n t a l an d n at i on al sec u r i t y su r v eil l a n ce. C om pu t er m et h ods an d prog ra m s i n bi om edi c in e. 2010; 100(1) : 16 - 23.
Original Paper
Loading high-quality paper...
Comments & Academic Discussion
Loading comments...
Leave a Comment