ࡱ > b d a bjbjUU > ? ? ) % % % % % #% #% #% 8 [% g% #% x {% % % % % % N % % kx mx mx mx mx mx mx z } J mx % & % % & & mx % % % % x & & & & % % % % kx & & kx & & N T X % @ #% & U 0 Wx x 0 x V } & } ` X X 4
} % b ` & & & & & & & & mx mx & & & & x & & & & } & & & & & & & & & # : Resumo
Sistemas de recomendao so uma popular ferramenta para sugerir produtos, servios e informao para potenciais consumidores, que se baseiam em transaes passadas e feedback de outros usurios tm interesses comuns. Com o enorme crescimento de usurios, produtos e informaes disponveis na internet e o rpido surgimento de novos servios online, a tarefa de gerar diversas recomendaes por segundo, para milhares de usurios se tornou uma necessidade e um desafio. Muitos sistemas de recomendao sugerem itens para os usurios utilizando tcnicas de filtro colaborativo, que processam o histrico de itens vistos, comprados ou avaliados pelos usurios. Dois principais problemas enfrentados pela maior parte das abordagens de filtro colaborativo so a escalabilidade e a esparsidade da matriz de perfil dos usurios, que foram superados com sucesso pela tcnica de modelos de fatores latentes. As concepes mais bem sucedidas de modelos de fatores latentes so baseadas em fatorao de matrizes. Dentre os algoritmos de fatorao de matrizes, o mnimos quadrados alternantes (ALS, do ingls alternating least squares), se destaca pelo fato de que suas etapas de clculo so facilmente paralelizveis. Neste trabalho propomos uma metodologia para comparao do desempenho entre duas implementaes paralelas do algoritmo ALS, um executado com o paradigma MapReduce no arcabouo Apache Hadoop, e outro executado no arcabouo Apache Spark, que faz uso de blocos de forma cuidadosa para reduzir o processo de coleta de lixo da JVM e tambm para melhor utilizar operaes de lgebra linear de alto nvel. So realizados experimentos de avaliao de desempenho quanto acurcia das recomendaes e ao tempo de treinamento de ambos os algoritmos quando executados em diferentes conjuntos de dados publicamente disponveis, de diferentes tamanhos e pertencentes a diferentes domnios de recomendao. Os resultados experimentais confirmam que a implementao no Spark mais eficiente, uma vez que o processamento realizado em memria e no em disco como no Hadoop.
Palavras-chave: sistemas de recomendao escalveis, filtro colaborativo, fatorao de matrizes, alternating least squares, Apache Spark, Apache Hadoop, MLlib, Mahout.
Abstract
Recommender systems are now a popular tool used to suggest products, services and information to potential consumers, based on their profile of past transactions and feedback from other users that share similar interests. With the tremendous growth of users, products and information made available on the web and the rapid introduction of new e-business services, performing many recommendations per second for millions of users has become a necessity and a challenge. Many recommender systems suggest items to users employing collaborative filtering techniques, which process historical records of items that the users have viewed, purchased, or rated. Two major problems that most collaborative filtering approaches have to resolve are scalability and sparseness of the users profile matrix, which have been successfully overcome with the use of latent factor models technique. The most successful realizations of latent factor models are based on matrix factorization. Among the algorithms for matrix factorization, alternating least squares (ALS) stands out because its computations are easily parallelizable. In this work we propose a methodology for comparing the performance of two parallel implementations of the ALS algorithm, one executed with MapReduce in Apache Hadoop framework and another executed in Apache Spark framework, which makes careful use of blocking to reduce JVM garbage collection overhead and to utilize higher-level linear algebra operations. We perform experiments to evaluate the accuracy of generated recommendations and the execution time of both algorithms, using publicly available datasets with different sizes and from different recommendation domains. Experimental results show that running the recommendation algorithm on Spark framework is in fact more efficient, once it provides in-memory processing, in contrast to Hadoops two-stage disk-based MapReduce paradigm.
Keywords: scalable recommender systems, collaborative filtering, matrix factorization, alternating least squares, Apache Spark, Apache Hadoop, MLlib, Mahout.
PAGE \* MERGEFORMAT2
^ d R k D E F G
ͻxtxtxtxtxtxkxtdt h?Ln hY5O hY5O mH nH uhY5O j hY5O UhY hY5O 5\mH sH BhY hY5O B*CJ OJ QJ ^J aJ fH mH ph%%% q
sH hY hY5O mH sH "hY hY5O 5CJ \aJ mH sH h)I hY5O 5\ hB hY5O 5\ hB hY5O 6] hB hY5O hB hY5O 5CJ \aJ " F G d gdD $dh ha$gdV
$dh a$gd5
$dh a$gd5 $dh `a$gd{y $a$gdS
$dh a$gdU<
$dh a$gdU< $
+ dh `a$gdB $dh ha$gdA
$dh ha$gdV dh gd?Ln $a$ d gdD < P 1h:p . A!5"5#n$n% Dp j3 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 2 0 @ P ` p 2 ( 0 @ P ` p 0 @ P ` p 0 @ P ` p 0 @ P ` p 0 @ P ` p 0 @ P ` p 8 X V x OJ PJ QJ _HmHnHsHtH R ` R Gvn N o r m a l d CJ OJ QJ _HaJ mHsHtH X X d!n 0 H e a d i n g 1 $$dh x @& 5CJ PJ \aJ n n x9 0 H e a d i n g 2 7 $$
& F
h @dh x @&^@`
5PJ \n n x9 0 H e a d i n g 3 7 $$
& F
h 0dh x @&^`0
5PJ \n n x9 0 H e a d i n g 4 7 $$
& F
h `dh x @&^``
5PJ \x x t @ 0 H e a d i n g 5 1 $$
& F
h @&^` B*OJ PJ QJ ^J ph$?` ~ ~ t @ 0 H e a d i n g 6 1 $$
& F
h @&^` 6B*OJ PJ QJ ]^J ph$?` ~ ~ t @ 0 H e a d i n g 7 1 $$
& F
h @&^` 6B*OJ PJ QJ ]^J ph@@@ t @ 0 H e a d i n g 8 1 $$
& F
h ` @&^`` ! B*CJ OJ PJ QJ ^J aJ ph@@@ t @ 0 H e a d i n g 9 1 $$
& F
h 0 @&^0` ' 6B*CJ OJ PJ QJ ]^J aJ ph@@@ D A D 0 D e f a u l t P a r a g r a p h F o n t R i R 0 T a b l e N o r m a l 4
l 4 a ( k (
0 N o L i s t N / N d!n 0 H e a d i n g 1 C h a r 5CJ OJ QJ \^J aJ N / N x9 0 H e a d i n g 2 C h a r 5CJ OJ QJ \^J aJ N / N x9 0 H e a d i n g 3 C h a r 5CJ OJ QJ \^J aJ N / !N x9 0 H e a d i n g 4 C h a r 5CJ OJ QJ \^J aJ J / 1J t @ 0 H e a d i n g 5 C h a r B*OJ QJ ^J ph$?` P / AP t @ 0 H e a d i n g 6 C h a r 6B*OJ QJ ]^J ph$?` P / QP t @ 0 H e a d i n g 7 C h a r 6B*OJ QJ ]^J ph@@@ R / aR t @ 0 H e a d i n g 8 C h a r B*CJ OJ QJ ^J aJ ph@@@ X / qX t @ 0 H e a d i n g 9 C h a r # 6B*CJ OJ QJ ]^J aJ ph@@@ > @ > D 0 H e a d e r
8!d . / . D 0 H e a d e r C h a r > > D 0 F o o t e r
8!d . / . D 0 F o o t e r C h a r R R D 0 B a l l o o n T e x t d CJ OJ QJ ^J aJ N / N D 0 B a l l o o n T e x t C h a r CJ OJ QJ ^J aJ :
: &~z 0 T O C H e a d i n g @& tH> >
14 0 T O C 2
p&
d ^ 2 2
= 0 T O C 1
&
d 6 U 6 A 0 H y p e r l i n k >*B*ph F F
14 0 T O C 3 "
(&
d ^ PJ tH@ 2@ t @ 0 L i s t P a r a g r a p h
# ^ 4 4 ; 0 B i b l i o g r a p h y $ B QB r 0 a p p l e - c o n v e r t e d - s p a c e B ' aB , 0 C o m m e n t R e f e r e n c e CJ aJ B rB ( , 0 C o m m e n t T e x t ' d CJ aJ N / N ' , 0 C o m m e n t T e x t C h a r CJ OJ QJ ^J aJ @ j qr@ * , 0 C o m m e n t S u b j e c t ) 5\F /F ) , 0 C o m m e n t S u b j e c t C h a r 5\B B I| 0 P l a c e h o l d e r T e x t B*ph > " > u 0 C a p t i o n , d 5CJ \aJ P P A 0
N o S p a c i n g - CJ OJ QJ _HaJ mHsHtH @ # @ ,? 0 T a b l e o f F i g u r e s . M^ 0
T a b l e G r i d I :V/ 0 a k
/ CJ ^J aJ M^ 0
L i g h t S h a d i n g Y:V0 0 a j;@
j;
jD jD k
44 0 f
f
A B*CJ ^J aJ ph 5\ 5\ 5\ 5\ M^ 0
M e d i u m L i s t 2 =:V1 0 a j;@
j;
jQ
jQ
jQ
jQ
j
j
k
44 1 . B*CJ OJ PJ QJ ^J aJ ph
CJ aJ , /!, o E s t i l o 1 2 F PK ! [Content_Types].xmlj0Eжr(Iw},-j4 wP-t#bΙ{UTU^hd}㨫)*1P' ^W0)T9<l#$yi} ;~@(Hu* Dנz/0ǰ $ X3aZ,D0j~3߶b~i>3\`?/[G\!-Rk.sԻ..a濭? PK ! ֧ 6 _rels/.relsj0 }Q%v/C/} (h"O
= C?hv=Ʌ%[xp{۵_Pѣ<1H0ORBdJE4b$q_ 6LR7`0̞O,En7Lib/Se е PK ! ky theme/theme/themeManager.xml M
@}w7c(Ebˮ CAǠҟ7՛K
Y,
e.|,H ,lxɴIsQ}#Ր ֵ+!,^$j=GW)E+&
8 PK ! P theme/theme/theme1.xmlYOo6w toc'vuر-Mn iP@I}úama[إ4:l ЯGRX^6؊>$ !)O^rC$y@ /yH*)UDb`}"qۋJחX^)I`nEp)liV[]1M<OP6r=zgbIguSe bORD۫qu gZo~ٺlAplxpT0+[}`j zA V2Fi@qv֬5\|ʜ̭NleXd sjcs7f
W+Ն7` g ȘJj|h(KD-
dXiJ؇(x$( :;˹! I_TS 1?E ??ZBΪmU/ ?~xY'y5g&/ɋ>GMGeD3Vq%'#q$8K )fw 9:ĵ
x}rxwr:\TZaG*y8IjbRc|XŻǿI
u3KGnD 1NIBs
RuK>V.E L+M2#'fi ~V vl{u8z H
* :( W ☕
~JTe\O*tHGHY }KNP*ݾ˦ TѼ9/#A7qZ $*c?qUnwN%Oi4 =3ڗP
1Pm \\9Mؓ2aD];Yt\[x]}Wr|]g-
eW
)6-rC Sj
id DЇAΜIqbJ#x꺃 6k#ASh&ʌt(Q% p%m&]caS l=X\P1Mh9MVdDAaVB[݈fJíP|8 քAV^f
Hn- "d>znNJ ة>b&2vKyϼD:,AGm\nziÙ .uχYC6OMf3or$5N HT[XF64T,ќM0E)`#5XY`פ ;%1U٥m;R>QD DcpU'&LE/pm%]8firS4d 7y\`JnίI R 3U~7+#m qBiDi*L69mY&iHE=(K&N!V.KeLDĕ{D vEꦚdeNƟe(MN9ߜR6&3(a/DUz<{ˊYȳV)9Z[4^n5!J?Q3eBoCM m<.vp IYfZY_p[=al-Y}Nc͙ŋ4vfavl'SA8|*u{-ߟ0%M07% < ҍ PK !
ѐ ' theme/theme/_rels/themeManager.xml.relsM
0wooӺ&݈Э5
6?$Q
,.aic21h:qm@RN;d`o7gK(M&$R(.1r'JЊT8V"AȻHu}|$b{ P8g/ ]QAsم(#L[ PK- ! [Content_Types].xmlPK- ! ֧ 6 + _rels/.relsPK- ! ky theme/theme/themeManager.xmlPK- ! P theme/theme/theme1.xmlPK- !
ѐ ' theme/theme/_rels/themeManager.xml.relsPK ]
( ( ( ( ( +
" $ + !
_ T o c 4 2 1 7 2 4 4 2 7
_ T o c 4 2 1 7 2 4 4 2 8 _ G o B a c k &
&
uC0* %, gE6 +)> !+ ~ E
36; 4e4 -0 &8YB wC9 . z@* JDB Wx>FJ :NG% 5hč xG;iऌ |8j%K Gl^KU :n^ k>Mo% =0w2z IT~M@l -^-`OJ QJ ^J o( ^`OJ QJ ^J o( o ^ `OJ QJ ^J o( ^ `OJ QJ ^J o( m^m`OJ QJ ^J o( o =^=`OJ QJ ^J o(
^
`OJ QJ ^J o( ^`OJ QJ ^J o( o ^`OJ QJ ^J o( ^`o(
C h a p t e r . ^` . pL^p`L . @ ^@ ` . ^` . L^`L . ^` . ^` . PL^P`L . ^`OJ QJ ^J o( e^e`OJ QJ ^J o( o 5 ^5 `OJ QJ ^J o( ^`OJ QJ ^J o( ^`OJ QJ ^J o( o ^`OJ QJ ^J o( u^u`OJ QJ ^J o( E^E`OJ QJ ^J o( o ^ `OJ QJ ^J o( ^`o( C h a p t e r - ^` . pL^p`L . @ ^@ ` . ^` . L^`L . ^` . ^` . PL^P`L . -^-`OJ PJ QJ o( ^`OJ QJ ^J o( o ^ `OJ QJ ^J o( ^ `OJ QJ ^J o( m^m`OJ QJ ^J o( o =^=`OJ QJ ^J o(
^
`OJ QJ ^J o( ^`OJ QJ ^J o( o ^`OJ QJ ^J o( ^`OJ QJ ^J o( ^`OJ QJ ^J o( o p^p`OJ QJ ^J o( @ ^@ `OJ QJ ^J o( ^`OJ QJ ^J o( o ^`OJ QJ ^J o( ^`OJ QJ ^J o( ^`OJ QJ ^J o( o P^P`OJ QJ ^J o( ^`OJ PJ QJ o( e^e`OJ QJ ^J o( o 5 ^5 `OJ QJ ^J o( ^`OJ QJ ^J o( ^`OJ QJ ^J o( o ^`OJ QJ ^J o( u^u`OJ QJ ^J o( E^E`OJ QJ ^J o( o ^ `OJ QJ ^J o( ^`o(
C h a p t e r . ^` . pL^p`L . @ ^@ ` . ^` . L^`L . ^` . ^` . PL^P`L . ^ `OJ QJ ^J o( ^`OJ QJ ^J o( o ^`OJ QJ ^J o( {^{`OJ QJ ^J o( K ^K `OJ QJ ^J o( o ^`OJ QJ ^J o( ^`OJ QJ ^J o( ^`OJ QJ ^J o( o ^`OJ QJ ^J o( -^-`OJ QJ ^J o( ^`OJ QJ ^J o( o p^p`OJ QJ ^J o( @ ^@ `OJ QJ ^J o( ^`OJ QJ ^J o( o ^`OJ QJ ^J o( ^`OJ QJ ^J o( ^`OJ QJ ^J o( o P^P`OJ QJ ^J o( e ^e` o( C h a p t e r - ^` . pL^p`L . @ ^@ ` . ^` . L^`L . ^` . ^` . PL^P`L . # ^`56 CJ OJ QJ \] ^J aJ o( C h a p t e r - ^` . pL^p`L . @ ^@ ` . ^` . L^`L . ^` . ^` . PL^P`L . P^`P @^@` . 0^`0 . . `^`` . . . ^` . . . . ^` . . . . .
^`
. . . . . .
`^`` . . . . . . .
0^0` . . . . . . . . e ^e` o( C h a p t e r - ^` . pL^p`L . @ ^@ ` . ^` . L^`L . ^` . ^` . PL^P`L . -^-`OJ PJ QJ o( ^`OJ QJ ^J o( o ^ `OJ QJ ^J o( ^ `OJ QJ ^J o( m^m`OJ QJ ^J o( o =^=`OJ QJ ^J o(
^
`OJ QJ ^J o( ^`OJ QJ ^J o( o ^`OJ QJ ^J o( ,^,`OJ QJ ^J o( ^`OJ QJ ^J o( o ^ `OJ QJ ^J o( ^ `OJ QJ ^J o( l^l`OJ QJ ^J o( o <^<`OJ QJ ^J o( ^ `OJ QJ ^J o( ^`OJ QJ ^J o( o ^`OJ QJ ^J o( ,^,` . ^` . L^ `L . ^ ` . l^l` . <L^<`L . ^ ` . ^` . L^`L . ^`o( C h a p t e r - ^` . pL^p`L . @ ^@ ` . ^` . L^`L . ^` . ^` . PL^P`L . P^`P @^@` . 0^`0 . . `^`` . . . ^` . . . . ^` . . . . .
^`
. . . . . .
`^`` . . . . . . .
0^0` . . . . . . . . P^`Po( C h a p t e r @^@` . 0^`0 . . `^`` . . . ^` . . . . ^` . . . . .
^`
. . . . . .
`^`` . . . . . . .
0^0` . . . . . . . . -^-`OJ QJ ^J o( ^`OJ QJ ^J o( o ^ `OJ QJ ^J o( ^ `OJ QJ ^J o( m^m`OJ QJ ^J o( o =^=`OJ QJ ^J o(
^
`OJ QJ ^J o( ^`OJ QJ ^J o( o ^`OJ QJ ^J o( :n +) &8 % JD 5h Wx>F =0w :NG k>Mo xG;i wC9 4e4 z@ E
3 |8j uC IT~ !+ Gl =0w gE Hd e * Hd ^ L^ ^ * Rt Iw , Jw , Kw , Lw , Mw , Nw , Ow , Pw , Qw , Rw , Sw , Tw , Uw , Vw , Ww , Xw , Yw , Zw , [w , \w , ]w ,w , < K ^w , _w , `w , aw , bw , cw , dw , ew , fw , gw , hw , iw ,w , jw , kw , lw , mw , nw , ow , pw , qw , rw , sw , tw , uw , vw , ww , xw , yw , zw , {w , |w , }w , ~w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w ,]w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w ,w , - w ,w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , w , x , x , R `9 :? l 3- 3 6 A U =Z \ l j5 V c = ET [ ] q + R5 C Xm ` ( 6 FG \ a ~l y s@ @ mI J yQ \ u| " Z z $ k7 < d` w
B
>
H
`P
o
J LU $c b A* ; Y^ f m
*
OO
j vp Y Z f 4 EA F n r { # H5 X [ \ 4 dV t T} > ? 4 O =^ q r s K[ l} :! 7 E YZ ][ \ ~r a}
W ( 2 : _ m 7 !9 = E Ef a
& + J
3 F L Y "y {0 ? +A ER &X c e h# D: Um | @ ! L1 ! ! %! xx! )" G" x" # 9# # 32# P6# 6# iL# #g# Or# R,$ U% F% "% ,)% B% RD% E% & (& & T& V& r' ' ,' *' 2' v@' S' 0t' q( ( I`( s( A
) A) d) *3* c?* ^* ;b* Wz* + #+ v-+ q6+ =+ :J+ J+ L+ `R+ n+ 0~+ v, 0, 8, -_, o, w, - - C- P- _- or- w- D. V. @{. 1/ @L/ rR/ b/ ~/ ;Z0 )1 11 hD1 vD1 m1 |1 2 T@2 A2 CT2 y2 f&3 )3 ,3 >3 g3 4 "4 e-4 .4 A4 Y4 X 5 5 1#5 B25 d5 :y5 #6 Wc6 4f6 |6 -|6 Y7 Vg7 v7 7 '8 z=8 A8 \8 a8 u8 9 N19 A9 |M9 4h9 x9 : # : $: ^;: @: pG: Y: u: +; ; +; D,; A; J; L; d; y; 7< I< U< ~< = = / = 2= "> )> K> E[> OB? _? _? d? Zv? t @ @ A 4A `A B B B GB mMB PB .\B xB C KC D (D -D !2D aD ~D _E JE &]E rE btE oF 9F JF PF BWF G ?G H 0H /pH CI I )I &dI UgI |I @J NJ WJ 2_J K K K Q^K zmK L :L P;L zBL VFL KL SL cL lL 9 M [M N N $N 9N BN fN TmN doN {N Y5O OO WO p^O bnO P 5"P /P <7P V7P BP BP JP zP Q fQ vQ OR .S 1S :S D_S NbS cS _|S &
T T U T s2T E9T xT U !U A/U 7U 'SU 8lU
V )V (V TV \V gV W W @W )W =-W -GW X BX oNX ;PX yX ~ Y 1Y EDY +cY lY SmY qY Z -Z xTZ 2tZ 8[ yi[ n[ .t[ O?\ m]\ ]
] %] 5] }O] U] mx] y] 0^ ^ i ^ M^ ]^ h^ j^ _ +_ B_ F_ ed_ *m_
` I` J` {e` a =%a h)a -a I8a Sa ma
b Jb /b w3b 7b ;b zy {y z Bz &~z F{ sk{ Rq{ B| 6| .| I| Q| } 0} D}
U} X} Z ~ >~ O~ i[~ l~ * : i vj = @8 G iO X u R " 14 f k % ? \ fk q c / , 8 :
K > ,? G f o pu < s 5 )G c h ? C -` x `/ Tc p r{ ` g iz ?" @ E -K 5
V ,\ 3 < g@ % 8 ,= ] _ Ix & }" G* AR s\ ! o o Y - ) 8x ^ 9 " B mC I N O g / F* =/ _7 c , F G S w` u k + , E I / ZO KS 2y qX 2: = d c= ,[ ^ !v B$ +
3 8C 'h Uu : 3 Y In } *M i n | 6 8B c {q
Q $ , 07 Z a l \ i. K2 5 = = sT d Xu | H 1 .1 (C dP U. 6 _? X :[ l ? u > C 5M :g } 5 7 B H
O V bd 2j y >& I `h Nz h( ( /3 E [ b Ze i B 7
g i 9w 0 M ,O @W k m s y X% , a eh 7 hH J K { $ \ f ,p ~ 1
M j1 Y % }) 8 ; Q< LO T Y s F 4P Xc J < A 0E xH X
Y _ z A 1J > W $ . m 0 < E &o A9 c o +
, '$ & ` c d k 0o
6 / N Q h h
H! I %5 ; aA Z a R & 8 N ] Sm A D J $T rr ; j l r~
A A i r D
i
}u ?} : % &' S/ !; X ){ { " [U %n D J V o H ]{ ~ f D E MM lh t> m# ' O W b m | > S 2 D * ; X0 KV V HZ M J f p W 7 %9 D< LF f @ v = _u Uv s' - ? .D d Jf k e 1 s* c+ "5 X k P- ? ? Z w X B< G mH P % nF <[ $i m u ' - A i ! Z [ , : t@ T c -m s 0 $, G\ p e = {b z U6 NN q
+ 1 &J M xU \ S vE L kN P Q i m * O _\ #t y \ 5 5 K 8 ? @ J 9K ys ! & - H [V ] C . ( ] b m r %X "q r z @ @ @ U n k n o w n G * Ax T i m e s N e w R o m a n 5 S y m b o l 3. * Cx A r i a l 7. @ C a l i b r i 7 @ C a m b r i a 5. . [` ) T a h o m a ?= * Cx C o u r i e r N e w ; W i n g d i n g s A C a m b r i a M a t h " 9g9g8' C 5n 0 $P D ! x x R e s u m o D a n i a H e l i o `
2
2 Oh +'0
< H T
` l t | Resumo Dania Normal_Wordconv Helio 2 Microsoft Office Outlook @ F# @
i71@ a@ a C ՜. +,0 h p |
Resumo Title
! " # $ % & ' ( ) * + , - . / 0 1 2 3 4 5 6 7 8 9 : ; < = > ? @ A B C D E F G H I J K L M N O P R S T U V W X Z [ \ ] ^ _ ` c R o o t E n t r y F ݿe 1 T a b l e .~ W o r d D o c u m e n t > S u m m a r y I n f o r m a t i o n ( Q D o c u m e n t S u m m a r y I n f o r m a t i o n 8 Y C o m p O b j y
F' Microsoft Office Word 97-2003 Document
MSWordDoc Word.Document.8 9q
F# Documento do Microsoft Office Word
MSWordDoc Word.Document.8 9q