Sei sulla pagina 1di 20

Jinguang Liu & Roopa Datla

Final Project: Research Paper

12/21/13

Table of Contents
WEB USAGE MINING ............................................................................................................... 2

BACKGROUND AND MOTIVATION...................................................................................... 2

WHAT IS WEB MINING?.......................................................................................................... 2

WH WEB USAGE MINING?................................................................................................... !

HOW TO "ER#ORM WEB USAGE MINING?....................................................................... !

WEB MINING A""$ICATIONS ............................................................................................. %&

SUMMAR ................................................................................................................................ %'

RE#ERENCES .......................................................................................................................... 2(

Course: CS5 5

!nstructor: Dr" #ang

Jinguang Liu & Roopa Datla

Final Project: Research Paper

12/21/13

Web Usa)e M*n*n)


++ "atte,n D*s-o.e,/ an0 *ts a11l*-at*ons

Background and Motivation


$ith the e%plosi&e gro'th o( in(or)ation sources a&aila*le on the $orl+ $i+e $e* an+ the rapi+l, increasing pace o( a+option to !nternet co))erce- the !nternet has e&ol&e+ into a gol+ )ine that contains or +,na)icall, generates in(or)ation that is *ene(icial to ./*usinesses" 0 'e* site is the )ost +irect lin1 a co)pan, has to its current an+ potential custo)ers" 2he co)panies can stu+, &isitor3s acti&ities through 'e* anal,sis- an+ (in+ the patterns in the &isitor3s *eha&ior" 2hese rich results ,iel+e+ *, 'e* anal,sis- 'hen couple+ 'ith co)pan, +ata 'arehouses- o((er great opportunities (or the near (uture"

What is Web Mining?


$e* )ining can *e *roa+l, +e(ine+ as +isco&er, an+ anal,sis o( use(ul in(or)ation (ro) the $orl+ $i+e $e*" 4ase+ on the +i((erent e)phasis an+ +i((erent 'a,s to o*tain in(or)ation- 'e* )ining can *e +i&i+e+ into t'o )ajor parts: $e* Contents 5ining an+ $e* 6sage 5ining" $e* Contents 5ining can *e +escri*e+ as the auto)atic search an+ retrie&al o( in(or)ation an+ resources a&aila*le (ro) )illions o( sites an+ on/line +ata*ases though search engines / 'e* spi+ers" $e* 6sage 5ining can *e +escri*e+ as the +isco&er, an+ anal,sis o( user access patterns- through the )ining o( log (iles an+ associate+ +ata (ro) a particular $e* site"

Course: CS5 5

!nstructor: Dr" #ang

Jinguang Liu & Roopa Datla

Final Project: Research Paper

12/21/13

Why Web Usage Mining?


!n this paper- 'e 'ill e)phasi7e on $e* usage )ining" Reasons are &er, si)ple: $ith the e%plosion o( ./co))erce- the 'a, co)panies are +oing *usinesses has *een change+" ./co))erce- )ainl, characteri7e+ *, electronic transactions through !nternet- has pro&i+e+ us a cost/e((icient an+ e((ecti&e 'a, o( +oing *usiness" 2he gro'th o( so)e ./*usinesses is astonishing- consi+ering ho' ./co))erce has )a+e 0)a7on"co) *eco)e the so/calle+ 8on/line $al/5art9" 6n(ortunatel,- to )ost co)panies- 'e* is nothing )ore than a place 'here transactions ta1e place" 2he, +i+ not reali7e that as )illions o( &isitors interact +ail, 'ith $e* sites aroun+ the 'orl+- )assi&e a)ounts o( +ata are *eing generate+" 0n+ the, also +i+ not reali7e that this in(or)ation coul+ *e &er, precious to the co)pan, in the (iel+s o( un+erstan+ing custo)er *eha&ior- i)pro&ing custo)er ser&ices an+ relationship- launching target )ar1eting ca)paigns- )easuring the success o( )ar1eting e((orts- an+ so on"

How to perform Web Usage Mining?


$e* usage )ining is achie&e+ (irst *, reporting &isitors tra((ic in(or)ation *ase+ on $e* ser&er log (iles an+ other source o( tra((ic +ata :as +iscusse+ *elo';" $e* ser&er log (iles 'ere use+ initiall, *, the 'e*)asters an+ s,ste) a+)inistrators (or the purposes o( 8ho' )uch tra((ic the, are getting- ho' )an, re<uests (ail- an+ 'hat 1in+ o( errors are *eing generate+9- etc" =o'e&er- $e* ser&er log (iles can also recor+ an+ trace the &isitors3 on/line *eha&iors" For e%a)ple- a(ter so)e *asic tra((ic anal,sis- the log (iles can help us ans'er <uestions such as 8(ro) 'hat search engine are &isitors co)ing> $hat pages are the )ost an+ least popular> $hich *ro'sers an+ operating s,ste)s are )ost co))onl, use+ *, &isitors>9
Course: CS5 5 !nstructor: Dr" #ang 3

Jinguang Liu & Roopa Datla

Final Project: Research Paper

12/21/13

$e* log (ile is one 'a, to collect $e* tra((ic +ata" 2he other 'a, is to 8sni((9 2CP/!P pac1ets as the, cross the net'or1- an+ to 8plug in9 to each $e* ser&er"

0(ter the $e* tra((ic +ata is o*taine+- it )a, *e co)*ine+ 'ith other relational +ata*ases- o&er 'hich the +ata )ining techni<ues are i)ple)ente+" 2hrough so)e +ata )ining techni<ues such as association rules- path anal,sis- se<uential anal,sis- clustering an+ classi(ication- &isitors3 *eha&ior patterns are (oun+ an+ interprete+"

2he a*o&e is the *rie( e%planation o( ho' $e* usage is +one" 5ost sophisticate+ s,ste)s an+ techni<ues (or +isco&er, an+ anal,sis o( patterns can *e place+ into t'o )ain categories- Pattern 0nal,sis 2ools an+ Pattern Disco&er, 2ools- as +iscusse+ *elo' in +etail"

Pattern Analysis Tools $e* site a+)inistrators are e%tre)el, intereste+ in <uestions li1e ?=o' are people using the site>? ?$hich pages are *eing accesse+ )ost (re<uentl,>?- etc" 2hese <uestions re<uire the anal,sis o( the structure o( h,perlin1s as 'ell as the contents o( the pages" 2he en+ pro+ucts o( such anal,sis )ight inclu+e: 1" the (re<uenc, o( &isits per +ocu)ent2" )ost recent &isit per +ocu)ent3" 'ho is &isiting 'hich +ocu)ents@" (re<uenc, o( use o( each h,perlin1- an+ 5" )ost recent use o( each h,perlin1" 2he techni<ues o( $e* usage patterns +isco&er,- such as association- path anal,sisse<uential patterns- etc" :'ill *e illustrate+ *elo' in +etail"

Course: CS5 5

!nstructor: Dr" #ang

Jinguang Liu & Roopa Datla

Final Project: Research Paper

12/21/13

2he co))on techni<ues use+ (or pattern anal,sis are &isuali7ation techni<ues- AL0P techni<ues- Data & Bno'le+ge Cuer,ing- an+ 6sa*ilit, 0nal,sis" =o'e&er- this paper )ainl, (ocuses on the Pattern Disco&eries- an+ the Pattern 0nal,sis 'ill not *e +iscusse+ (urther in +etail"

Pattern Discovery Tools Pattern Disco&er, 2ools i)ple)ent techni<ues (ro) +ata )ining- ps,cholog,- an+ in(or)ation theor, on the $e* tra((ic +ata collecte+"

Data Pre-processing Portions o( $e* usage +ata e%ist in sources as +i&erse as $e* ser&er logs- re(erral logsregistration/(iles an+ in+e% ser&er logs" 2his in(or)ation nee+s to *e integrate+ to (or) a co)plete +ata set (or +ata )ining" =o'e&er- *e(ore the integration o( the +ata- $e* log (iles nee+ to *e cleane+/(iltere+- using techni<ues li1e (iltering the ra' +ata to eli)inate outliers an+/or irrele&ant ite)s- grouping in+i&i+ual page accesses into se)antic units"

Filtering the ra' +ata to eli)inate irrele&ant ite)s is i)portant (or 'e* tra((ic anal,sis" .li)ination o( irrele&ant ite)s can *e acco)plishe+ *, chec1ing the su((i% o( the 6RL na)e- 'hich tells ,ou 'hat (or)at these 1in+ o( (iles are" For e%a)ple- the e)*e++e+ graphics can *e (iltere+ out (ro) the $e* log (ile- 'hose su((i% is usuall, the (or) o( 8gi(9- 8jpeg9- 8jpg9- 8D!F9- 8JP.D9- 8JPD9- can *e re)o&e+"

2he ne%t step is to integrate +ata (ro) all sources to (or) a &isitor pro(ile +ata" Ar 'e can sa,- the +ata in registration (iles :)ainl, &isitorsE +e)ographic an+ househol+

Course: CS5 5

!nstructor: Dr" #ang

Jinguang Liu & Roopa Datla

Final Project: Research Paper

12/21/13

in(or)ation; can *e appen+e+ to log an+ (or)s +ata" 2he (igure gi&es an e%a)ple o( +ata integration"

Pattern Discovery Techniques Converting IP addresses to Domain Names .&er, &isitor to a $e* site connects to the !nternet through an !P a++ress :(or e%a)ple1 F"22G"55"153;" .&er, !P a++ress has a correspon+ing +o)ain na)e- an+ these are lin1e+ through the Do)ain Ha)e S,ste) :DHS;" DHS can con&ert a +o)ain na)e that a &isitor entere+ in $e* *ro'ser into a correspon+ing !P a++ress" 0 &isitor3s !P a++ress can *e con&erte+ into a +o)ain na)e *, using the DHS s,ste) in re&erse- calle+ a re&erse DHS loo1up"

#ou can har+l, )ine an, 1no'le+ge )erel, (ro) an !P nu)*er" =o'e&er- i( ,ou con&ert the !P nu)*er into the +o)ain na)e- so)e 1no'le+ge can *e +isco&ere+" For e%a)ple,ou can esti)ate 'here &isitors li&e *, loo1ing at the e%tension o( each &isitor3s +o)ain na)e- such as "ca :Cana+a;I "au :0ustralia;I cn:China;- etc"

Course: CS5 5

!nstructor: Dr" #ang

Jinguang Liu & Roopa Datla

Final Project: Research Paper

12/21/13

Converting File Names to Page Titles 0 'ell/+esigne+ site 'ill ha&e a title :*et'een KtitleL an+ K/titleL; (or e&er, page" Rather than si)pl, report the (ile na)es :6RL; re<ueste+- a goo+ s,ste) shoul+ loo1 at these (iles an+ +eter)ine their titles" Page titles are )uch easier to rea+ than 6RLs- so a goo+ s,ste) shoul+ sho' page titles on reports in a++ition to 6RLs"

Path Analysis Draph )o+els are )ost co))onl, use+ (or Path 0nal,sis" !n the graph )o+els- a graph represents so)e relation +e(ine+ on $e* pages :or 'e*;- an+ each tree o( the graph represents a 'e* site" .ach no+e in the tree represents a 'e* page :ht)l +ocu)ent;- an+ e+ges *et'een trees represent the lin1s *et'een 'e* sites- 'hile the e+ges *et'een no+es insi+e a sa)e tree represent lin1s *et'een +ocu)ents at a 'e* site"

$hen path anal,sis is use+ on the site as a 'hole- this in(or)ation can o((er &alua*le insights a*out na&igational pro*le)s" .%a)ples o( in(or)ation that can *e +isco&ere+ through path anal,sis are:

GFM o( clients 'ho accesse+ /company/products/order.asp *, starting at /company an+ procee+ing through /company/whatsnew.html- an+ /company/products/sample.html I

JNM o( clients le(t the site a(ter (our or less page re(erences"

2he (irst rule tells us that GFM o( &isitors +eci+e+ to )a1e a purchase a(ter seeing the sa)ple o( the pro+ucts" 2he secon+ rule in+icates an attrition rate (or the site" Since

Course: CS5 5

!nstructor: Dr" #ang

Jinguang Liu & Roopa Datla

Final Project: Research Paper

12/21/13

)an, users +on3t *ro'se (urther than (our pages into the site- it is tact(ul to ensure that )ost i)portant in(or)ation :pro+uct sa)ple- (or e%a)ple; is containe+ 'ithin (our pages o( the co))on site entr, points"

Grouping 6sers usuall, can +ra' higher/le&el conclusions *, grouping si)ilar in(or)ation" For e%a)ple- grouping all Hetscape *ro'sers together an+ all 5icroso(t *ro'sers together 'ill sho' 'hich *ro'ser is )ore popular on the site- regar+less o( )inor &ersions" Si)ilarl,- grouping all re(erring 6RLs containing the 'or+ 8#ahoo9 sho's ho' )an, &isitors ca)e (ro) a #ahoo ser&er" For e%a)ple:

http://search",ahoo"co)/*in/search>pO$e*P5iners

Filtering Si)ple reporting nee+s re<uire onl, si)ple anal,sis s,ste)s" =o'e&er- as the co)pan,3s $e* *eco)es )ore integrate+ 'ith the other (unctionalit, o( the co)pan,- (or e%a)plecusto)er ser&ice- hu)an resources- )ar1eting acti&it,- anal,sis nee+ to rapi+l, e%pan+" For e%a)ple- the co)pan, launches a )ar1eting ca)paign" Print an+ tele&ision a+s no' are +esigne+ to +ri&e consu)ers to a $e* site- rather than to call an FNN nu)*er or to &isit a store" Conse<uentl,- trac1ing online )ar1eting ca)paign results is no longer a )inor issue *ut a )ajor )ar1eting concern"

A(ten it3s +i((icult to pre+ict 'hich &aria*les are critical until consi+era*le in(or)ation has *een capture+ an+ anal,7e+" Conse<uentl,- a $e* tra((ic anal,sis s,ste) shoul+ allo' precise (iltering an+ grouping in(or)ation e&en a(ter the +ata has *een collecte+"

Course: CS5 5

!nstructor: Dr" #ang

Jinguang Liu & Roopa Datla

Final Project: Research Paper

12/21/13

S,ste)s that (orce a co)pan, to pre+ict 'hich &aria*les are i)portant *e(ore capturing the +ata can lea+ to poor +ecisions *ecause the +ata 'ill *e s1e'e+ to'ar+ the e%pecte+ outco)e"

Filtering in(or)ation allo's a )anager to ans'er speci(ic <uestions a*out the site" For e%a)ple- (ilters can *e use+ to calculate ho' )an, &isitors a site recei&e+ this 'ee1 (ro) 5icroso(t" !n this e%a)ple- a (ilter is set (or 8this 'ee19- an+ (or &isitors that ha&e the 'or+ 85icroso(t9 in their +o)ain na)e :e"g"pro%,12")icroso(t"co);" 2his coul+ *e co)pare+ to o&erall tra((ic to +eter)ine 'hat percentage o( &isitor3s 'or1 (or 5icroso(t"

Dynamic Site Analysis / ignette StoryServer

2ra+itional $e* sites 'ere usuall, static =25L pages- o(ten han+/cra(te+ *, $e*)asters" 2o+a,- a nu)*er o( co)panies- inclu+ing Qignette an+ 5icroso(t- )a1e s,ste)s that allo' an =25L (ile to *e +,na)icall, create+ aroun+ a +ata*ase" 2his o((ers a+&antages li1e- inclu+e+ centrali7e+ storage- (le%i*ilit,- an+ &ersion control" 4ut it also presents pro*le)s (or so)e $e* tra((ic anal,sis *ecause the si)ple 6RLs nor)all, seen on $e* sites )a, *e replace+ *, &er, long lines o( para)eters an+ cr,ptic !D nu)*ers" !n such s,ste)s- <uer, strings t,picall, are use+ to a++ critical +ata to the en+ o( a 6RL :usuall, +eli)ite+ 'ith a 8>9;" For e%a)ple- the (ollo'ing re(erring 6RL is (ro) Hetscape Search:

http://search"netscape"co)/cgi/in/search>searchOFe+eralP2a%PReturnPFor)&cpOntserch

Course: CS5 5

!nstructor: Dr" #ang

Jinguang Liu & Roopa Datla

Final Project: Research Paper

12/21/13

4, loo1ing at the +ata a(ter the 8>9 'e see that this &isitor searche+ (or 8Fe+eral 2a% Return For)9 on Hetscape *e(ore co)ing to our site" Hetscape enco+es this in(or)ation 'ith a <uer, para)eter calle+ 8search9 an+ separates each search 1e,'or+ 'ith the 8P9 character" !n this e%a)ple- 8Fe+eral-9 82a%-9 ?Return? an+ 8For)9 each is re(erre+ to as para)eter &alues"

4, loo1ing at this in(or)ation- co)panies can tell 'hat the &isitor is loo1ing (or" 2his in(or)ation can *e use+ (or altering a $e* site to ensure that in(or)ation &isitors are loo1ing (or is rea+il, a&aila*le- an+ (or purchasing 1e,'or+s (ro) search engines"

Coo!ies

Coo1ies usuall, are ran+o)l, assigne+ !Ds that a $e* ser&er gi&es to a $e* *ro'ser the (irst ti)e that the *ro'ser connects to a $e* site" An su*se<uent &isits- the $e* *ro'ser sen+s the sa)e !D *ac1 to the $e* ser&er- e((ecti&el, telling the $e* site that a speci(ic user has returne+" Coo1ies are in+epen+ent o( !P a++resses- an+ 'or1 'ell on sites 'ith a su*stantial nu)*er o( &isitors (ro) !SPs" 0uthenticate+ userna)es e&en )ore accuratel, i+enti(, in+i&i+uals- *ut the, re<uire each user to enter a uni<ue userna)e an+ pass'or+so)ething that )ost $e* sites are un'illing to )an+ate" Coo1ies *ene(it $e* site +e&elopers *, )ore easil, i+enti(,ing in+i&i+ual &isitors- 'hich results in a greater un+erstan+ing o( ho' the site is use+" Coo1ies also *ene(it &isitors *, allo'ing $e* sites to recogni7e repeat &isits"

For e%a)ple- 0)a7on"co) uses coo1ies to ena*le their 8one/clic19 or+ering s,ste)" Since 0)a7on alrea+, has ,our )ailing a++ress an+ cre+it car+ on (ile- ,ou +on3t re/

Course: CS5 5

!nstructor: Dr" #ang

1N

Jinguang Liu & Roopa Datla

Final Project: Research Paper

12/21/13

enter this in(or)ation- )a1ing the transaction (aster an+ easier" 2he coo1ie +oes not contain this )ailing or cre+it car+ in(or)ationI that in(or)ation t,picall, 'as collecte+ 'hen the &isitor entere+ it into a (or) on the $e* site" 2he coo1ie )erel, con(ir)s that the sa)e co)puter is *ac1 +uring the ne%t site &isit"

!( a $e* site uses coo1ies- in(or)ation 'ill appear in the coo1ie (iel+ o( the log (ile- an+ can *e use+ *, a $e* tra((ic anal,sis so(t'are to +o a *etter jo* o( trac1ing repeat &isitors"

6n(ortunatel,- coo1ies re)ain a )isun+erstoo+ an+ contro&ersial topic" 0 coo1ie is not an e%ecuta*le progra)- so it can3t (or)at ,our har+ +ri&e or steal pri&ate in(or)ation" 5o+ern *ro'sers ha&e the a*ilit, to turn coo1ie processing on or o((- so users 'ho chose not to accept the) are acco))o+ate+"

Association "ules !)ple)ent association rules to on/line shopper can generall, (in+ out his/her spen+ing ha*its on so)e relate+ pro+ucts" For e%a)ple- i( a transaction o( an on/line shopper consists o( a set o( ite)s- 'hile each ite) has a separate 6RL" 2hen the shopper3s *u,ing pattern 'ill *e recor+e+ in the log (ile- an+ the 1no'le+ge )ine+ (ro) 'hich- can *e the (or) li1e the (ollo'ing:

3NM o( clients 'ho accesse+ the 'e* page 'ith 6RL /co)pan,/pro+ucts/*rea+"ht)lalso accesse+ /co)pan,/pro+ucts/)il1"ht)"

Course: CS5 5

!nstructor: Dr" #ang

11

Jinguang Liu & Roopa Datla

Final Project: Research Paper

12/21/13

@NM o( clients 'ho accesse+ /co)pan,/announce)ents/special"ht)l- place+ an online or+er in /co)pan,/pro+ucts/pro+ucts1"ht)l

0nother e%a)ple o( association rule sho'n *elo' is the lin1e+ associations *et'een online pro+ucts an+ search 1e,'or+s" !t )easures the association *et'een the 1e,'or+s use+ to search an+ the +i((erent pro+ucts actuall, sol+" 2his (or) o( report can also *e achie&e+ *, Dynamic Site Analysis / Vignette StoryServer )entione+ a*o&e"

Se#uential Patterns Se<uential patterns +isco&er, is to (in+ the inter/transaction patterns such that the presence o( a set o( ite)s is (ollo'e+ *, another ite) in the ti)e/sta)p or+ere+ transaction set" $e* log (iles can recor+ a set o( transactions in ti)e se<uence" !( the 'e*/*ase+ co)panies can +isco&er the se<uential patterns o( the &isitors- the co)panies can pre+ict users3 &isit patterns an+ target )ar1et on a group o( users" 2he se<uential patterns can *e +isco&ere+ as the (ollo'ing (or):

Course: CS5 5

!nstructor: Dr" #ang

12

Jinguang Liu & Roopa Datla

Final Project: Research Paper

12/21/13

5NM o( client 'ho *ought ite)s in /pc'orl+/co)puters/- also place+ an or+er online in /pc'orl+/accessories/ 'ithin 15 +a,s

Clustering Clustering i+enti(ies &isitors 'ho share co))on characteristics" 0(ter ,ou get the custo)ers3/&isitors3 pro(iles- ,ou can speci(, ho' )an, clusters to i+enti(, 'ithin a group o( pro(iles- an+ then tr, to (in+ the set o( clusters that *est represents the )ost pro(iles"

4esi+es in(or)ation (ro) $e* log (iles- custo)er pro(iles o(ten nee+ to *e o*taine+ (ro) an on/line sur&e, (or) 'hen the transaction occurs" For e%a)ple- ,ou )a, *e as1e+ to ans'er the <uestions li1e age- gen+er- e)ail account- )ailing a++ress- ho**ies- etc" 2hose +ata 'ill *e store+ in the co)pan,3s custo)er pro(ile +ata*ase- an+ 'ill *e use+ (or (uture +ata )ining purpose" 0n e%a)ple o( clustering coul+ *e:

5NM o( clients 'ho applie+ +isco&er platinu) car+ in /+isco&ercar+/custo)erSer&ice/ne'car+- 'ere in the 25/3N age group- 'ith annual inco)e *et'een R@N-NNN S 5N-NNN"

Clustering o( client in(or)ation can *e use+ on the +e&elop)ent an+ e%ecution o( (uture )ar1eting strategies- online an+/or o((/line- such as auto)ate+ )ailing ca)paign"

Decision Trees 0 +ecision tree is essentiall, a (lo' chart o( <uestions or +ata points that ulti)atel, lea+s to a +ecision" For e%a)ple- a car/*u,ing +ecision tree )ight start *, as1ing 'hether ,ou

Course: CS5 5

!nstructor: Dr" #ang

13

Jinguang Liu & Roopa Datla

Final Project: Research Paper

12/21/13

'ant a 1

or 2NNN )o+el ,ear car- then as1 'hat t,pe o( car- then as1 'hether ,ou

pre(er po'er or econo),- an+ so on" 6lti)atel, it can +eter)ine 'hat )ight *e the *est car (or ,ou"

Decision trees s,ste)s are incorporate+ in pro+uct/selection s,ste)s o((ere+ *, )an, &en+ors" 2he, are great (or situations in 'hich a &isitor co)es to a $e* site 'ith a particular nee+" 4ut once the +ecision has *een )a+e- the ans'ers to the <uestions contri*ute little to targeting or personali7ation o( that &isitor in the (uture"

Web Mining Applications


$e* )ining e%ten+s anal,sis )uch (urther *, co)*ining other corporate in(or)ation 'ith $e* tra((ic +ata" 2his allo's accounting- custo)er pro(ile- in&entor,- an+ +e)ographic in(or)ation to *e correlate+ 'ith $e* *ro'sing- 'hich ans'ers co)ple% <uestions such as: A( the people 'ho hit our $e* site- ho' )an, purchase+ so)ething> $hich a+&ertising ca)paigns resulte+ in the )ost purchases- not just hits> Do ), $e* &isitors (it a certain pro(ile> Can ! use this (or seg)enting ), )ar1et>

Practical applications o( $e* )ining technolog, are a*un+ant- an+ are *, no )eans the li)it to this technolog," $e* )ining tools can *e e%ten+e+ an+ progra))e+ to ans'er al)ost an, <uestion"

$e* )ining can pro&i+e co)panies )anagerial insight into &isitor pro(iles- 'hich help top )anage)ent ta1e strategic actions accor+ingl," 0lso- the co)pan, can o*tain so)e su*jecti&e )easure)ents through $e* 5ining on the e((ecti&eness o( their )ar1eting
1@

Course: CS5 5

!nstructor: Dr" #ang

Jinguang Liu & Roopa Datla

Final Project: Research Paper

12/21/13

ca)paign or )ar1eting research- 'hich 'ill help the *usiness to i)pro&e an+ align their )ar1eting strategies ti)el,"

For e%a)ple- the co)pan, )a, ha&e a list o( goals as (ollo'ing: !ncrease a&erage page &ie's per sessionI !ncrease a&erage pro(it per chec1outI Decrease pro+ucts returne+I !ncrease nu)*er o( re(erre+ custo)ersI !ncrease *ran+ a'arenessI !ncrease retention rate :such as nu)*er o( &isitors that ha&e returne+ 'ithin 3N +a,s;I Re+uce clic1s/to/close:a&erage page &ie's to acco)plish a purchase or o*tain +esire+ in(or)ation;I !ncrease con&ersion rate :chec1outs per &isit;"

2he co)pan, can i+enti(, the strength an+ 'ea1ness o( its 'e* )ar1eting ca)paign through $e* 5ining- an+ then )a1e strategic a+just)ents- o*tain the (ee+*ac1 (ro) $e* 5ining again to see the i)pro&e)ent" 2his proce+ure is an on/going continuous process"

He%t- 'e 'ill gi&e so)e e%a)ples on $e* 5ining applications"

Measuring Return of Online Advertising Campaigns 0s online a+&ertising *anners *eco)e )ore popular- co)panies using the) accuratel, )easure o&erall return on a+&ertising in&est)ent" 2his *ene(its *oth a+&ertisers an+ sites

Course: CS5 5

!nstructor: Dr" #ang

15

Jinguang Liu & Roopa Datla

Final Project: Research Paper

12/21/13

running a+s *ecause it allo's a+&ertising rates to &ar, accor+ing to their success" Proper )easure)ent o( a+&ertising reports centers on t'o speci(ic areas:

23ant*t/: =o' )an, i)pressions 'ere +eli&ere+ (or each a+ *anner an+ page- an+ ho' )an, people clic1e+ on each a+> 2hese are usuall, reporte+ as impressions an+ clickthroughs"

23al*t/: A( people 'ho clic1e+ on an a+ *anner- ho' )an, actuall, purchase+> 2his return is *est )easure+ *, su*tracting a+&ertising e%penses (ro) the resulting re&enue"

For co)panies o((ering a+ space on their site- reporting a+ i)pressions an+ clic1/through rates (or an, page running a+&ertise)ents is i)portant" For co)panies running *anner a+s on other sites- prospect <ualit, can *e )easure+" 0 )anager shoul+ e&aluate *oth the e((ecti&eness o( in+i&i+ual a+ *anners an+ the e((ecti&eness o( each $e* page 'ith an a+" 4, co)*ining these- an a+&ertiser opti)i7es his or her a+&ertising *, selecting the *est co)*ination o( a+ *anner an+ $e* page (or a++itional a+ place)ents" 2he (ollo'ing report gi&es an e%a)ple o( a car site an+ the )ost e((ecti&e a+s (or each page" H*)4est Cl*-5+T4,o3)4 Rates fo, Ea-4 "a)e
,ate "a)e Na6e Front Page/+e(ault"ht) A0 Na6e 5ustang Se*ring Cor&ette !ntrigue Ca)aro Classi(ie+s/class"ht)l Cor&ette !ntrigue =ot 2opics/hotne's"asp 5ustang I61,ess*ons 3@-1NN 3@-JNN 2-1NN J@-1NN 3-GNN -FNN 1N-NNN 3-@NN Cl*-5+T4,o3)4s 21-NN 1-@NN 3-1NN 2-1NN 1-5NN 3NN 2NN 1-2NN Cl*-5 T4,o3)4 Rate J"2M @"NM 3"@M 3"3M 1"JM 3"1M 2"NM 35"3M Cost R3-@1N R3-@JN R -21N RJ-@1N R -3GN R FN R1-NNN R3@N

Course: CS5 5

!nstructor: Dr" #ang

1J

Jinguang Liu & Roopa Datla

Final Project: Research Paper

12/21/13

Cor&ette Se*ring Ca)aro

3-3NN 5- NN J-2NN

2NN 2NN 2NN

J"1M 3"@M 3"2M

R33N R5 N RJ2N

Measuring Return of E-Mail Campaigns Co)*ining $e* tra((ic tools 'ith an e/)ail )erging progra) is one o( the *est 'a,s to )a%i)i7e return on )ar1eting e/)ails" Custo) 6RLs or <uer, strings are assigne+ to each prospect" $hen the prospect rea+s the )essage an+ clic1s the 6RL- the $e* tra((ic anal,sis progra) +eter)ines 'ho the &isitor is an+ *egins the appropriate sales process" For e%a)ple: http://'''"co)pan,"co)/+e(ault"ht)>QisitorOe/)ailTa++ress

$hen this 6RL is clic1e+- the uni<ue i+enti(ier (e-mail_address; is passe+ to the $e* site" 4, using a (ilter *ase+ on the <uer, string- it is possi*le to )easure the *est lea+s" 4, lin1ing the e/)ailTa++ress to a custo)er in(or)ation +ata*ase- sales personnel can recei&e reports sho'ing contact na)es- phone nu)*ers- an+ )ore" Results )easure+ in +ollars can also *e calculate+ *, lin1ing to a )ar1eting +ata*ase"

Mar et !egmentation $hen co)*ine+ 'ith a pro(iling s,ste)- $e* )ining can per(or) )ar1et seg)entation" 2his allo's $e* )ar1eters to target ca)paigns an+ )essages to speci(ic groups" For e%a)ple- an online )usic co)pan, using a pro(iling s,ste) coul+ create reports sho'ing the +i((erences in *ro'sing *eha&ior *ase+ on age ranges" 2he, )ight (in+ that )ost o( their actual purchasers are in their 2N3s" 0n un+erstan+ing o( 'hat in(or)ation 'as attracti&e to other &isitors 'oul+ *e &alua*le in +esigning a $e* site to appeal to a 'i+er

Course: CS5 5

!nstructor: Dr" #ang

1G

Jinguang Liu & Roopa Datla

Final Project: Research Paper

12/21/13

au+ience" 2his in(or)ation coul+ *e use+ to e%pan+ content an+ +irect &isitors to the right place" For e%a)ple: Na6e A)e Ran)e Re73ests

Age Ranges per Page


S4o8s t4e a)e ,an)es9 *n %( /ea, *n-,e6ents9 of .*s*to,s to ea-4 1a)e. "a)e Na6e A)e Ran)e Re73ests Ho6e "a)e N: 1 1N: 1 3 2N: 2 23 3N: 3 13 @N: @ 11 5N: 5 F ",o03-t "a)e N: 1 1N: 1 @ 2N: 2 2J3 3N: 3 1@1 @N: @ 23 5N: 5 G1 C3sto6e, S311o,t "a)e 2N: 2 21 3N: 3 1@ @N: @ F 5N: 5 F

2he a*o&e just a (e' sa)ple applications o( $e* )ining" 0s 'e sai+ *e(ore- the practical applications o( $e* )ining are a*un+ant" $e* )ining is not e%clusi&el, i)ple)ente+ in the !nternet- it can also i)ple)ente+ in !ntranet :a)ong the users 'ithin the co)pan,)ainl, e)plo,ees; an+ .%tranet :suppliers an+ custo)ers 'ith .D! connection;" $ith the $e* )ining on !ntranet an+ .%tranet- co)pan, can achie&e resource opti)i7ation 'ithin the organi7ation- an+ i)pro&e custo)er ser&ice an+/or suppl, train )anage)ent 'ith the suppliers :upstrea); as 'ell as 'ith the custo)ers :+o'nstrea);"

Course: CS5 5

!nstructor: Dr" #ang

1F

Jinguang Liu & Roopa Datla

Final Project: Research Paper

12/21/13

Summary
!t is a re&olution that the !nternet has gro'n (ro) a si)ple search tool to a gol+ )ine" Co)panies (in+ a ne' an+ *etter 'a, to +o *usiness: ./co))erce through the !nternet" =o'e&er- ./*usiness cannot just *uil+ a 'e* site an+ then sit *ac1 an+ reap the *ene(its'hich- in )ost cases- is (ruitless" Co)panies ha&e to i)ple)ent $e* )ining s,ste)s to un+erstan+ their custo)ersE pro(iles- an+ to i+enti(, their o'n strength an+ 'ea1ness o( their ./)ar1eting e((orts on the 'e* through continuous i)pro&e)ents" !nternet is a gol+ )ine- *ut onl, (or those co)panies 'ho reali7e the i)portance o( $e* )ining an+ a+opt a $e* )ining strateg, no'"

Course: CS5 5

!nstructor: Dr" #ang

Jinguang Liu & Roopa Datla

Final Project: Research Paper

12/21/13

eferences

Intelligence or !usiness at "-Speed- *, Ste&e Russell http://'''"+)re&ie'"co)/e+itorial/+)re&ie'/printTaction"c()>.+!DO1 GF Data#ase Access $ver the %e#: .%ten+ing the $ire- *, Dr" Larr, R" =arris http://'''"+)re&ie'"co)/)aster"c()>Ha&!DO2 Data &ining and the %e#' %hat (hey )an Do (ogether- *, 5ar, Dar&in http://'''"+)re&ie'"co)/e+itorial/+)re&ie'/printTaction"c()>.+!DO@2N Data &ining on the %e#* *, Don R" Dreening http://'''"'e*techni<ues"co)/archi&es/2NNN/N1/greating/ %e# &ining' in ormation and +attern Discovery on the %%% *, Ro*ert Coole,- 4a)sha+ 5o*asher- Jai+eep Sri&asta&a http://'''/users"cs"u)n"e+u/U)o*asher/'e*)iner/sur&e,/sur&e,"ht)l (he ,ive %ire %e# Data &ining %hite +aper http://'''"l'**s"co)/'hitepaper"ht)l Integrating and &ining %e# Data in -our %arehouse- *, Jesus 5ena http://'''"+)re&ie'"co)/e+itorial/+)re&ie'/printTaction"c()>.+!DO1@N2 Analy.ing -our $nline )ustomer- *, Larr, 4ohn http://'''"+)re&ie'"co)/e+itorial/+)re&ie'/printTaction"c()>.+!DO1G@2 ,everaging the Internet to /educe the (ime and )ost o Doing !usiness - *, Dena 4auc1)an http://'''"+)re&ie'"co)/e+itorial/+)re&ie'/printTaction"c()>.+!DOFG %e# &ining %hite paper - Driving !usiness Decisions in %e# (ime http://'''"accrue"co)

Course: CS5 5

!nstructor: Dr" #ang

2N

Potrebbero piacerti anche