A Comparative Study On Feature Selection in Text Categorization
A Comparative Study On Feature Selection in Text Categorization
-/0 122$34$
[)GXGIB$[KRUT]OFHV\WcTZb&FHY [·aHTVE[)]OTHv
Q@W[wb\XxSaHTb_TT&bUN"[wWXVUTZRUTZx´TZRUVUD@N x@FCTZx@p ]H]Hb VUFC[)TZFCTV\x@T4FKY@WXJ~Y@TZbUR\Bb4TGXQ@VUBZN xSVUTWXRUxwTTGXV\WXWXWXN"YFHY R\WXYD@R\[)WXbD@R\WXVUbUF)TR\Y2b\Q@D@x@J8FHVUdCGXFOhHx@N²T$TZ[Kx´BER\D"D@R\F>T_F)J;JI[)Q@BE¢&GXD@GiD@WXTaHWXBEaOF D T p
B$x@[KFCg@BQ@Q@G¸N/[)V\Td,YCR=FKJ8JV\T[±¥CQ@x@TFCY@BBQ@d N"¦;TZY»§R/ZBh)FHWXYCV\P@J8FOQ@VUb·N>v [KRU WXFHD@YTZbU]CT±[)WXBY>VUWX¦©RUUTZª_VUW¸§[hONw[KVUQ Tp P@RUªQSQ@WXVUaHR\TTTZxwYbUPRU[.[KDSBRUT&T&VE[)WcRUYWXD@Y@J8FOFHWXY@bUV\TN>]RUB[)TFOR\V\WXVUN"FHP@Y4Q@b]Cb·¢&h)[)D@J;WXFHYiFHV b\h)T[)T$YS[KWXYCBExJ8DwFOV\VUQ@TNN"Y@WX[)F)¥CRUaOQ@WXTFOT&xY&RUJ8T]V\V\[)FHNÌWXNY_¢¢&R\D@T[)TMbBFHJ8GXT$TN [bUbpp
RUbUQkR\VU[)TGiYSWX]HYCRUJ;D¯FHVU¦ N [)A@R\§ZWcv FOY>¦;^U§hk[ ¬» b\R[KRUWXbUR\WcB/¦8L®4U§Zh[KY@x RUTZVUN RUWXY@DkB[)GXY2Q@x@bUFHTZbN/RUT D@P@TV\TTb\x@RUWXTN>R\T[KVUN/RUWXFHWcYSY_TFKxwJRURUD@DSV\T&TBbUDSFHY@FHGXxSxiWcR\v WXFH YD@[KT&GBPSFHVUN/FHgkP@[)Q@g@REWX[)GXWXRURUWXWXFOTY b
F)RUWXJ FO[_Y@b B·[)WXY"RUTZR\]HD@FHT4V\d"x@T']O¹WcaOY@TWXRUY WXFH[_YR\v T VUN±[email protected]@P@[)Y@VUFOx/gRU[)D@gSTWcGXTWXReYCd_R\VUTFHb\PCRUWXd/N>B[KFHRUN/WXFHP@Y2Q@DR[K[ bp
| =79 : "!U)$ #%=&: '(*)=)+ ),-e6¸:=(< ./1032 [2WwRU¦ZWXY[N/X T&§=BFH¢&N/D@TZP@VUGXT\TCXµWcRed2Wcb=F)R\&J D@WwT&YCN¦ XÀQ@Nw§&g[)Y@TZV»x FK[2JRUbUVEPk[)[)WXY@BT_WXY@BZ]FHN"x@FCPSBGcQ@TZN/WXTRed2YCRUF)b·J h
¢&.D@FCWXBBEQ@D N"[¯TZYR\R TVUJ8V\NT¥CQ@FCTBY@BQ@BdV\b$v£WXb2°9RUD@T±TYCBFOQ@N"NwP@gkQ@TR\VTx{FKJ RUx@D@FCT±BQSx@N"FCBTQ@YCN/RUb2TYCWXY R RU[)WXYSFO]x Y@b&YÌD[)WXbb4R\[ D@RUT2WXN/aHFCT4B$B[)FOgSN"Q@G¸P@[)GXVUTdCWXb\ReWcdZTHFKv ^J Ww D@Z¦ Y_TU/TZY§R\v VUFOPdÀBFHN/P@Q@R[ p
| *=?$- 6;:!)79?$·67: . 6 2 WXY@D@x@T TPk¬=T Y@bUxSRET[)YCR\WcR$b\vwRUWX°9B DT2[KbiBFH[&N/YP@[KQ@RUQ@R\TVEx[)GOJ;aHFH[)V4GXQ@T·T=[)BEF)D9JCB$T[)VUFR\TWº]HJFOE»VU[Kd±Y@RUx;D@< T [)V\T
bUVU^RETZ[)G½Q@[KR\R\RUWcQb\TRUxw[)WXG&B$[)[KWXPSYCGP@J;G¸[KFHGXWcY@VUB·N]H[)QRU[)WX[KFOR\]HWcY@FOT=bY´N"Î s WXFCbwh x@n [ TZh Gcn@GXBZWXl'VUY@WXÏ RU]vIT;F)V\JJ@WXFHFH¢YYSFHTVUBx2BFHFON/[)Y@bUN"bUb\WXFCx@FOBTZY@W¸VU[)GXb d>RUWXR\FOQSD@Y@bUT&Tb»Rex[K¢Y@WXF xY p bU[)REYS[)xR\Wcb\RURUDkWXB[)R2gTB$Re[)¢R\TT]HTFOYwVUdCT·[)h BE[)D_Y@Q@x¯Y@RUWX¥CD@Q@TY¯TRUBTFHV\NwNgSWXWcYYS[Tx9RUVRU[KD@WXY@TWXY@B$[K]&RUBTFH]OV\FHP@V\¬»Q@d bp
¢&¢&[·D@d´TV\TBFO{YCRUWcWXbY@]HRUTZDSY@T BYCd9Q@RNw[Kgkg@GXTTwVF)F)JJ=RU[WXN"R\TTVU^b N E [KEY@[)x Y@<xB[ÀF p B$FC[KBRUBTQ@]OV·FHh VU ]d <KWXb h bUPkTBWº¹B bUBFOVUTbF)J=T$[)BED R\TVUN WXYCRUF2Re¢F2bUBFOVUTb*
RUF)DSJITRUWXYCN/Q@TNw&b gk<TFVIBZF)BJ@Q@R\VUWcb&N/¢&T=WXRUR\D@D@T FHQ@EiR FCE$BhCBQ@[)V\Y@b»x ¢&X WXR\D@WXb FHQ@R\D@R,T&<)hRUFOR[KWXbiGYCYCQ@Q@NwNwgkgkTTVV ¬ ¦CE§&H >ABD L M 7¦ </>e§ ¬ G¦ E /</>Á§
F)gkJ4TRexS¢FTZBZTQ@Y1N"E4T[)YCY@R\b$x h=<.RUDSWXTbYx@T'R\¹D@Y@T"TNwx Q@R\FwR\Qg[)T GMWcYJ8FHV\N>[)R\WXFHY BV\WXRUTV\WcFOY ¬ A ¦CE§&H N>AB[)D : ¬ ¦GE /< > §/?
P@GXD@TT_CWXBZRedFHhON"bUPSWXN/Q@RWcG¸[K[KRUV WXFHRUY/FwF)^JÑ&L[)®.Y@&xbUUBªwFOVUv TbD[Kb [w¥CQ[Kx@V[KRUWXBBFHN p
LM G¦ E 1<§
G¦ E /<§&HGXFH]
L M ¦GE§ L M ¦N<§
j[Y@NFOVU[NEFO[)VGXWXx@WºT³Ix±TVUaHTZ[)Y@GXBQ@T2THh gT[KReY@¢x9TTD@Y9TY@LBZT®44[KY@aHx [KGc^QST&bWXb&[KVURUT/D[KBR FHN/¬= P[ WXb p
[)YSx>WXbTbUR\WXN>[)R\TxQ@bUWXY@] VRUDS[[email protected] Y@FH[)V\BN>V\FH[KbUGcb/WX$RU[KTRUV\WXN"FHYwbgSJ8VUFHT$V2[K¶[email protected]Àx@F)b¢&[)N/YT¬=¦©B$ B$[)[)Y R\TYS]HFwFOVUGcdCFOvÂY@]H®4TZV4F)¢gkTTaO[)TB V$p h
¦GE /<§ GXFH] ¦ O _§9¦X O § BRUDSQ@TVE[)BR\FOTYCGXdRUWXY@BFO]HN"TZY@PB[Kd2VUTR[)x2gSRUGcFTMR\Wcb=D@TGXWX]H¬»DC [email protected]\PRUFOVUWXP@g@Q@QSG¸RU[)WXR\FHTYxi§h)Wº¢&JÊD@[KYWXBEd D2BWcTbGXG@RUD@WXY T
B$WXb[KbU¶CT2Y@F)J8FO¢&V4YwGXF)YS¢{FHRJ8RUVUF2TZ¥gQST.TY@VUTBGXd±W¸[)RUg@TGXV\TN"J8FOb·Vv_GX®4F)¢ TZY@p J8BVUTHTZh=¥R\QSD@TT Y@B¬=dw R\bUTREVU[)N/RUWXb$b\Î RUÒZWXÏ B v
x@[±T¦CE ]HPk<GXTZFOY@§2gxS[)DTGMYC[)J8b"R$T$v [K[±RUQ@ YV\F[KT>RUQ@N"b\VET[)T$GXT[KG.BbUR\aHQ@Wc[)V\FOGXT´YiQ@h&T"R\D@¢FKT9J_T]HBFCTFHVUFCNwF¯x@g@Y@WºWXJ\TZY@bUE"Twb [)R\FKD@Y@JT x [¯B$<[KRURUT[)TV\V\]ON T FHV\WXYdWXY pp
-| , . C)7 + $)C:=< *)8.."/,2
bUPkTBWº¹B bUBFOVUTbF)J [2RUTV\N²WXYCRUF2Re¢F"[)GXRUTV\Y[)R\T¢Ñ[·dCb °ÂTWXVUGXNÌg@Q@b\V_RUVU[)TZYSY@x]HR\ACD>WXVUWXFHb R\FO¶CVUWcWXY]HWXYÎ nH[KnZGXÏ GcdJ;FHP@V_V\FHaOPkFFHB·b\[)Tg@x>Q@G¸[K[KY@VUxwd TV\aHT[)x@GXQ@QB[KR\RUWXTFHxwY9gCWXYd
G¦ E&§ H L M 7¦ </>Á§ C¦ E </>e§
A
>BD RURUTZF2RUR TZVUTRR\B$VU[KWXTRUaHT[)]OGeFHhiVUWX[K$Y@[Kx±RUWXFHG½[KY±RUTÎ n)V_ [Kh P@nHP@ZGXÏ WXv Tx´ D@gCWXd bÖN/[)TY@RUDS]F[)xwY@TZx±bURU°ÂWXNWXGX[)g@RUQ@TZbV
A C¦ E&§ H N>>AB[)D : C¦ E < > /§ ? RUGXWXTZ¶HVUTN GXd&WcRUN/FP[)P@FOPkVURET$[)[KY@ViBWXT/Yg×UB[)GXb\FHTb\xTGXdFHY¯p V\TD@G¸[)F)RU¢TZx@BÙ4FOx@N"FCN"BQSFON"Y@GXTd YCRU[ b·vIR\TeRiVUN QSbUTWXbb
bUWXN/D@T&WcG¸^[KV»IR\BFwFON"RUD@P@T Q@UREª[)R\WcBFOFOYN"DP@[)Q@b RE[&[)R\R\WcWXFON"YiT=v BFHN/P@GXTCWXRed4F)J WwZ¦ Y_U/§h
¢&RU[wDSD@R\T VFORe[KbU¢WcT"YSF&Wcb\YSxSWXN"]wFBZWXbUG¸Q@T[)N"R_VUWXFKReTd±YCJ RIx@¦¸aON"FCTBT·BQ@[)RUN"FObUQSVUTZbEVUY§iTR\x9WXb4biR\Q@[KF"gb\WXF)Y@x@aO]TT~V\WXRU[aHDST2R\T"D@x@VUBTZFCFObUBbUD@Q@WXY@FON"GcT"xTZaHv4Y[KR.× GX Q@PTZTw[)VUWXNF)V\b J
jWXb&bU¢R\VUT$FH[KYS¶]HYSGXTd"bUb/WX5Y F)4J_QSTNwY@Q@BTRUQkx±[)gCGd"WcYR\J8D@FHT_V\N>N>[)[KR\VUWX]OFHWcY9Yk[)WXbwG@P@RUV\DFH[KgR"[Kg@R\D@WXGXT WcR\WXbUTBZb&FHVUF)T J
RUTZVUN"b·h@[)b&B$[KYgT_b\TTY WXY/RUD@WXb&T¥CQ@WXaH[)GXTYCRMJ8FHV\N( BFHACFOYSR\VUY@x´Tx@Y@DWX]ORU[KWXRUFHGºD@JÑYkÙ¯FK[)JGiRUD@PS[/TZVUY,PFHgk[KWXWc[)bwVg@BFKWXGXFOJÑWXReN"V\dwTP@G¸RU[)QSDRURU[KTZTR x9x[wx@gR\FC[KTBbUVUQ@TN£xN"FCTZFHYBYBZR\bQ@R\VUD@]Hb_T WXaHWXYTTY±b\RURUWXDSRUN>DkT[)[Kb\RRUTTB WXx Rp
G¦ E /<,§ HGcFO] L M G¦ E P <§ I±GXFH] L M C¦ E§ FCPB[KBWcVIQ@FKV\bJx@WXY9WXbUR\R\WXD@Y@T2BRi¹gSVUQ@b\RRiV\DkT[)G¸[)GºJRUvwTZx_ÔIx@T"R FCB0Q@N/[)YST2x YCRUb$1h)g[KY@T"x [)EiY9gk[)TV\[&g@WXR\RUTVVU[KN±VUd h
SFHVG¦ E P <R\§ZThCVUN/V[KbVUT_¢&R\TWXVURUN/D b&[)¢&Y¡WcGXGITZD¥[·QkaO[)T_G{[2BD@FHWXY@]HxSD@WcTR\V&WXFHb\YB[KFHGÂV\T_P@RUVUDkFO[)gY[)gSBWcFHGXNWXRed p RUDSTY RUDST b\RUV\TY@]HR\D F)J=RUD@T RUTV\N²WXb&x@TE¹kY@Tx R\FwgT
N"L[)BZM FOVUFHYwb\RUb4TR\V\TN"VUN/b·v b FKD@J~T¢&b\WXBx@FHTZV\GcTdwb$hkxSWÓRU³IDSTTV\VUWcTEYSJ;]2FHVUJ8TOVUhT[)¥CV\Q@TTZY@Y@BFHdCRv BFHN/P[)VE[)g@GXT
G¦ E§&H L M ¦GE*451P E*460k§
3
| ¬» Q ·?$·6Á+ $6@ .! 6 2 RUDSD@TÂTFHRUTZY@VUTZN£bbUN"R\VUTZTYYSR\]HWXFHRUY@DÀTBxÌVUWXRUT$TZ[KVUVUWXGXFHWXTYV$vWXb&Ve[KRx@WXWXB$b9[)GXGXgdw[KbUx@TWºx³ÊTZVUFOTYYCR&x@J;FCVUFHBQ N p
Re¢D@TT TY ¬= E b\[)RY@[)x R\WXbU<&RUWX[KB&Y@N/xwT$B$[)[Kb\YwQ@gVUTT&b=BR\FOD@N"T&PG¸[)[KBEVU¶_Tx2F)JkRUF WcYSRUx@D@TT PkTY@x@x@TWXYSbUBR\VUT4WXg@gkQ T pp N"bUDkTZ[)YVURMTZx BGX¢Q@bUFOR\VUTx@VUb4WXY@[K]kVUhT_[)V\b\TbUG¸Q@[)N/R\TWXxiY@h]_[KRUY@Dkx[)RRUDx@[KFCR&BQ@RUTN"V\N"TZYb&R\b&WXY"¢&R\WXD@RUD"T_ND@T·[)[·YCa d p
RU`4WXFOb\WXYwY@]"¢&WcR\R\D@D"TwFOReY@¢T4F x@p ¢&TZ]H[·VUd"TZT BZFHFKYCJIRUJ8WXVUY@TZ]OTTx@Y@FHBNµd RUREF&[)g@EGXQ@T_x@FK]OJÑT4[/TC¬»RURUTV\V\TN N"TZE Y@[KTY@bUb·x v WXWXGXYCd4J8FOF)VUaHN>TZVU[KG¸RU[)WXP@aHPSTOWcv YS]4 [)D@VUWXbT·[&N"F)TJkR\VUD@TFCG¸[)x´R\TWXx_b_x@YSFCFHBRQSRN"[KbUT¶ YCp RUb\b=P[)TZV\BTWº¹V\BHTh=G¸[)WeRUv THWXaOvXhITGXWXd R
x@[)R\FCWXTFHbY@b·Y@v FOR4eY Q@b\R\T_D@WXWXb&YCbUJ8FHTY@V\N>b\TH[KhIRUWXWXR&FHY"WXb&[KbUgWXN"FOQ@WXG¸R4[)VMR\TRUVUF"N R\p D@B$T[KRUT]O9BFHV\Zd«VUWX[KRUTbUV\b\WXFFHBZYiW p h
BFH[wF QSB·p \R FC[)RUB<)TZB$h ]HQSFHV$V\h#WXd b& <)RUDShkWcbT ¢&R\YCD@D@Q@TTVUNw"T YCgkQ@ÌTNwVWXgkF)b&J=TRUV.RUD@WXT2F)N"J&YCTRUQ@b WXNwN"<.gTFCTZBRUV4BDSQ@TF)V\J b4E R\¢&WcFCN/BWXRUBTD@Q@b FOV\E b_Q@R [)¢&YSE$WX$h(x RUD% <p g@Q@R&xSWÓ³ITV\TYCR&J8V\FHNR\D@T_Uªwh@^[)Y@xRUDST ¬» bURE[)R\Wcb\RUWXBHv
WXWXb4b4R\R\D@D@TwTwYCRUFOQ@RNw[)GgYCTZV&Q@NwF)JgR\TZWcV4N/F)TJb4x@YSFCTBWXQ@RUD@N"TTZ;V Y< R\b$Y@hIFHR\V D@T2E FCRUTBV\BN QSVUp b$]Hh=FC[)FCYSx@(x Y@TX b\b jx@FCPBQ@[)N/VE[)TN"YCTZRwRUTbUWXVN"WXY"WXG¸[)RUV\DSWXT Red aHA/[)B$GXQ@[KTGXBb·Q@v G¸[) R\WcDFO[)YwRwWXbWXb$R\hD@D@TF)RU¢ D@V\TBbUGXFHD@b\FOT GXx>Re¢FOYF
N"T·[)bUQSVUTWXb&x@TE¹Y@TZx RUF2gT x@°9FCT"BQ@QSN/bUT TYCjRUbw8 Nw7 ÔMQ@h=b\RRUD@gkTT [·R\aHF´TVEgk[)T ]OT"BFOYCY@Q@bUNwWXx@gkTTV\TVxFKJ4[ VUV\TTG¸[KG¸[)RURUTTZxxÂx@PFC[)BWXQ V·p v
¬ G¦ E /<,§ H ¦ O _§ ´¦)X& ¾¦'% I(_§
O %w§ 9¦ O _*§ ±¦+ O %w§ N"TZYR\b&PTZV4x@FCBQ@N/TYCR&WXY>R\D@VUTZbUD@FOGcx±RUQSY@WXY@]v D[)R.Wcb·h@¢T
BRUVEFO[)N"WXY@P@WXQ@Y@R\]wTb\R\TD@R$T&hbURUWXV\N/d>WcxSG¸[KWÓ³IVUWXTReV\d&TbUYCBR_FOVURUTD@bV\TF)bUJIDS[)FHGXGXGCx@RUbD@FHT&Y x@FCR\D@BQST2N"bUWXTN"YCRUWXG¸b=[)WXV\YwWcRe[d g@V\T$[)¶ p TaHTYÀPFOWcYCRF)J @l$¼ v
aHFH[)GXx/GXQ@¢&TZbD@WXFKBJiDx@V\FCTBbUQ@Q@N"GXR\b4TZYWXRY P[2[KWcVUV\T$b$[Kh@bU[KFHY@Ykx"[)g@R\GXD@TTaHY"[)BEGXQ@D@TFCFOF)bUJ Tj RUD@ T7 R\Ô D@v VUT b\D@D Tp ÐP@ Q@2 dRwÐ×UFKFHG¸[KJ RUVUD/[]HTZbUBZdCÙG½bU[K¢R\bUTTb\N"WÓ¹kN"bTT$V"bU[KB·B$Y"[)[KGXT4RUYDkR\[)DFR[·G¸aOg[)T"V\FO]HRUR\T&DD@BFHRUG¸Q@D@[KbET bU[)bUWcWºY@YS¹xSP@B·bw[)Q@RUR&F)WXFOJ [)YwY@xSxP@WcN/VURUFODSTg@Y@T GXb\TFOWcN"FOQ@Y@R b·pb v
aHD@[)F)GX¢9Q@T¢FKTJGXGCj WcR= FO7 P@Ô±RUWXN"WXb4WXBETZD@x FObUR\TD@Y TPTZTPkVrJ8TFHV\V\WcN>N/[KTY@YCBR[KTGcWcGXYdChRU[KD@BTBRFO[KVUbUx@¶IWXY@v ]wj R\B F p FHF)JV&J8DST$Wc[K]ORUD@Q@TV\V·T Î nOb\m ThGXTnKB·R\ÏWcFOv=Yi°9hCJ;TVUFH¢&NÌ[)YCY@R&F2RUFwVUTZTZx@Q@[KBN"R\WcFOWXY@Y9T¦¸[)TGXCGBR\TD@PST_R4x@VUTZTZ]HN"VUF)TZTa b p
B]HFOFOVUVUx@WX$WX[KY@RU]WXFHRUYwFFOP@YwV\TbUaCTWXaOFHTQ@Vb_[KGiTZx@aHFC[)BGXQQ@[)N/R\WXTFHYCY@R b.BF)FOJ&GcGXTVUTBZR\RUVUWXFHWXTY@aHb·[)Î nHGn [)h Y@nHx±ZÏ B$hO[)RUR\D@T T p WXx@Y@Q@]BZRUbUWXREFH[)YiY@hx[)[KY@VUx¯x±FHbURUg@FOb\P9TVUaO¢TwFOVURUx@D@bET2§TERU³IFÀTBTR\CbRUVUFOTZY´N"R\TD@GXd9T[)[K]HBB]OQ@VUTVEb\[)bUBWXd±aHTwF)J&V\T [ p
J8jYCFOQ@VUNwN>7 gk[KÔ±TY@VBaHTH[)FKvGXJ»Q@R\TZVb4DS[)T_WXgkY@TBWXReY@FH¢N/]2TZx@P@TYFCQ@BREl$Q@[)t RUN/WXFOR\TFY"YCRUnOF)b·tJ v dWXA/TGXWXxb&b¥C[KQRU[)WXbex@JeVE[)[)BR\R\WcFHBVUd/WXYPRUTZD@V T p BYG¸[K[)RUb\WXbUFHWºYi¹ThCV.¢F)TaHY@TTV.TRUx±D@T [_Tb\YCB$[)RUWXG¸VU[)T_gSGcRET[)V\b\]HdTb\R.RUTbUN±P[Kv BTHv&@FOV&[email protected]@[)N"W p
( =CÉ @9 Å Ç i@
WX Y@2 D@]´WXÐb[/FH[KR\]HGcDGXGXF)FO¢&¶ g3b[)G [VEb\[)[)RUY@Y@V¶C[KxWXWX]HY@ÔiDC] ÔR p A@FKJ8JÑ[FHV\B·¢ÑK[)VURU[KT_TZVU]Hx2[ FH]HV\U WXGXTFHp bgk[KVU[)]Od/G@WXaHTBTaHG¸Y[)[KbUGXQb\[Wº[)¹xSR\TWcFV4FOBZYPSQ@VUFKN"F)JÊaCTPkYCWXxTR·V p v
| [-;?·6 H x@RUDSFCT>BQ@]ON/FCFTYCxSR&Y@TB$bU[)bwR\TFK]HJ4FOVUB$WX[)$R\[KTRU]HWXFOFHVUY d¯PkVT[)VrYSJ8FH¶V\WXN>Y@]À[)YS]HBWXaHTHTh@YWev TH[ vXhCxSN"FT·BZ[)Q@bUN"QSVUTWXYCY@R·] h
V¢&[KD@RUD@TY TZV[KR\P@DP@[)GXdCY9WcYSPk]TV2g@WXB$Y[)[KR\VUTdw]HFOBVUG¸d9[KbUbUPkWºT¹VeTZJ8VUFOb4VUNR\Fw[)Y@RUD@BTT"P@[KVUb_FOWXg@bGXTb\N´R[Kv Y@x[)V\x
¢&[)¦8ÔYSD@ÔixwWXBEA@RUD_=§[email protected]Î nOFHVUsZTQ@PSÏ R\v VUP@T Q@b\R D@TYCT4FKRUJiWXTY@RUxD@P@T[)QSb\b»RT4[R\b\Fb\dPb\gk[)RUFHTV\R\N/bUD"Tb aHbUWXdCTbBZb\RURU[TFHVEN/V=[)FKbY@J¶OWX¢b&Tx"FH[V\GXx@xWXbUFC¢R=BTF)Q@WXJiN/]OB·DT[)R\YCRb$Rp h Q@bUÔiTZN/ÔPA@[)ThYCVE[)R ZRUJ8R\TZFHFGcVdCvTB$@[)R\[)ACTN"WX]HN"FOPSWXVUGcG¸WXTO[)ThVUb$FHGXh dCPShI[KRUY@¶WXN"x±\WXD@TTbY@RUV\RUBT$D@T/[)TR\x@b>N>FCT[À[Kb2P@x@P@Y@FCWXFHY@BRQ@]MRUN"J8V\VUT$FOTZ[)YNR2R"[&¢[Kbx@FHVUFC[KxSB Y bp
T]OFHVUWXTb&¢&WXR\D«[2BZFHYC¹xSTY@BT_b\BFHV\TJ8FOV&T$[)BEDB$[)R\T]HFOVUdCv bUWXb4WXY@WX]OY±GXT4x@PkWXbUFHRUWXWXYCY@R&BZRUWcWXYFHY [2R\aHF TBZRUFHV&b\P[)BZTHv . N"D@TZT RUBD@[email protected]gR&[)b\b\TTx´Y@b\FOWcR\Y´WXaTZWXRe d p
L&[)b\[)bUR\WX]HT]HYSFOTVUxd9R\FVE[)RUY@D@¶CT WXY@_]±YST$WcY9[)V\T¶ bU3RRUVE[)WcWXb2Y@gWXY@[K]4bUTxSx¯FBZFHQ@YN"R\TD@YCTR\biB$RU[)FR\TR\]HD@FOT&VUWXWXYTb p
P@WXY@Q@]_R·v R\D@ T4DSb\T WcN/B$[)WXR\G½[KT]HVUWXFOReVUd4WXTFKb=JiF)T$J@[)R\BED@D/TY@bUTTWXY@]HTDCWXgk]ODFHVgkFHR\FV\b»R\[KD@VUT4T WX¢Y@TP@WXQS]HR$DChORU¢&TZx_D@Q@TVUb T p
P@BG¸GX[)WXBb\WXbURWº¹WcTYSV\x@b$TÎ l$PkoqTÏ Y@x@[KY@TYSx9BT bUFO[)N"bUb\TwQ@FON"RUD@PSTRUWXV FHY@VUTb]Ob\VUQ@Tb\BEbUD,WXFH[)Yb N"Y[)TZWXRUaHD@T2FCx@Ð&b·[·Î ZdHÏ T§Zb v
"#$%'&()
bUgkTZFHBRURUDµWXFHYk¶ §3h2RUD@T¯[)Y@g@x VUT·Ôi[)¶Ôip ASTZhaHTY FHQ@PR\FOPWcTZYCVeR J8FOaHVU[KN"GXQ@WXY@Tb]¢[)GXTGwVUT R\D@HTÂmO¼ FHR\D@J8FOTVV BFOD@VUTPkFH V[.TQ@WXY_R\TR\VUTb2CY@R»TB·¢&[)RUb2TZ]HbUR\FHFHV\V\WXd9$[)R\BWcFHFOGXY_GXTBVUTZR\WcbUFOT$Y´[KVUBEWXb_D´BZÎ FHl$N"o h N/l h FHnSY@l GXd/h l$Q@Ò bUh ToZx Ï
bUbUdCdCb\NwRUTgkN/FHGXb"WXB_TVUaHQS[KGcGcT2Qk[)GXT$RU[)TZx¨V\Y@WXFOY@YÂ]RUgCD@d TÀ b[) fN/f T 7 B FOGcGX¦ TOBZtHRU¼WXFHYi§ZÎ h oZÏ WchiYSABGX° Q@x@j WXY@f] p GXRUTFOv bUP@b WXB_R\DSDG¸T[)[)VUYgkTTDk[)GXb$[)VUvT GºJ °9nSF)l JTh R\Q@D@mObUT_t TZx´x@x@FCFCFOBBY@Q@Q@GXdwN/N/RUTT[email protected]_T2DWXY±x@[·aHFCRUT_BD@Q@T2DCN"Q@J8NQSTZYGcG=[)R\b&Y BFORU[)GXDGcb\TZ[KbUBR4WX]HRUWXDYSFH[KTY xx
D
-*
P@ACl V\GXTF¦ TsO[)PSHBEWc¼D YS]§Q@Î l\7bUÏ WXY@[KPk]¨Y@Tx9V\L RULb v ® m¦ sHj ÒO¦ sH¼ Oj §¼ Î oZ§Ï 7Îhil$[)m'¦ YSÏ sHh_x¾OWX¼ Y@[/x@§ReÎQ@dCl$BZÒZP@RUÏ WXWXhiB$aH[)[/T9GMx@GXWcT$YTZ[KJ8BFHVUWXbUY@V\WXN>WXFHY@Y9[)]ÂR\[)WXgCFHP dY p bUP[)TZRw[KRVUGXRUT$FKWX[KR\J WcbUFO RwY"h ÒSFOWXl$Y@bt TbUWX[)RUN"Y@FHx9PSWXG¸Wc[)BO[/V=hRURUxSTF2b\WcaCRRUDkWXx@b\[)TTR&Rx¯TFKN/VJ [KP@Y@o GXx@h F)ÒHFHdHÒHN/Tn xwGXx@d´WXY´FCWXBYCÎQ@l\RUN/ÏF9hOTg@[±YCQ@RURUR&b$VEv [)x@WXWº Y@³IWXD@TY@VUWX] bb
VUFHTZR\RUD@VUTWXTV±aH[KaHG4[)V\[)W½[KP@RUPSWXVUFHFY[KBEF)DJ"YRUDS[KN"T T x TZQ@ RUFTBZV\b´BDSBZWcF¯FHGXGX¦ TsHBmORU¼ WXFO§YÎ oZÏ¢&v D@T4YVUT¯RU[)D@Y T p J8V\FHN Î l$oZÏ ¢&DSF9Q@bUTRUDST"J;Q@GXG&BFHGXGXTBR\WcFOYWXY@BGXQ@xSWcYS]´Q@YSG½[ p
RU¶ VE\[)WXY@WXDY@[)].b&bUT[ R&g@[)VUY@T$x2[K¶ RUp DSTT&aOTRUYTb\PR FObUWXTYR&R[)F)V\JTOPknH[)¼ VUR\WX¢&RUWXD@FHWXY@BETZD"xWXx@bWºRU³ID@TTVUTZbY[KR\N"GXdT h
[)bR\D@TwVUTb\Q@GXR_F)JY@TQ@VE[)GY@TRe¢FHVU¶Cb·Î nSl\Ï hi[)Y@x9ÔiÔA@D[Kb_[
.+/1032465'798:2;<'=+2>?;@2)8:=<'AB0DCE0F@GH2)5'7JIK2:LM+2)8:=NPORQSTUCV<'=
F;FB<W5'FBX5'2YMZ8:AL[ZEZM+M\^] __4`4K4 a 8:2=+2F8:bZDa F@M+Ma bAGY_)c5'246<'=a
gkF)J TGXHGcTZt x9v Ò x@¢FCFOBQ@VUx@N"bTZ¢&YR\WXbRUD± vbURE[) Y@D@xTw[KVUbUx±R\FHx@V\WcTTZaCb_W¸[)DR\[·WcaHFOT"Y [/@l N/v Ò T$v[)Y±°9GXT2TY@B]HFHR\Y D p F)J=VUTB·[)GXG»[KY@xP@VUTBZWcb\WXFHY [)b&PkTVeJ;FHVUN[)Y@BZT N/T$[)b\Q@VUTZb
bURUWXDSx@T TZVURUTVEx±[)WXY@RUD@WXY@T ]±HnbUTB$R·[)v R\T ]HD@FOVUTWXb\TT bB$RU[KDkRU[)TR ]OFH[KVUP@WXPkTb/T$[)BVF)[)aHR4TZV"GXT$RU[KFObUP@RWXBFHbwYSBbUT_QSBWXDY VUTZB$[)GXG H B$[)R\RUTFH]HREFO[)VUGIWXB$T[Kb RUJ8TFH]OQ@FHYSV\x>WcTZb4[)YSBFOx>VUVUBZTZFHBVUR V\TBR
°ÂB$[)[KbRUD@BTWXFO]OGXT2N"FHVUbUN/WXFHTb·N/FhkxSTwRUWcD@R\x@WXT_TFCb$N/BhSQ@WcT$N/YC[)R\Y/TTYCVUWXTRUb&b_b\RFHDY@VE[·GX[)daOR\TwTl b$Q@v h=n)P9 [)Y@RUB$F«xÀ[)R\J8J8TFHFO]HQ@VUFOTV\VUWXRUWX]HTTY±TZb4Y,TPkCT[)BEVb\DbUx@WX[)]HFCYSYSB]HTQ THx p v P@V\TBWXbUWXFOY H B$[KRUTR\]OFHFHR[KV\WcGITZB$b&[)J8R\FOTQ@]HY@FOx>VUWXT[Kb Y@J8xFHQ@BFHYSV\x VUTBZR
N"B$[KTZRUYT]OR·FHv VU d9D@R\TF J8V\B$T[K¥CRUQ@TT]OY@FHBZVUdwd F)JIFBZBQ@VU V\ThY@BJ8FOT V_aHT[)@VUWX[)TN/b=]OP@VUGXTHT$[KhiRU[)GXPSd_PJ;T$VUFH[KNVUb ¢&[)V\D@T_T[)V\TgkF)×UaHB$T [K[2RUT]O]OFHWXaHV\TWcTZY/bwbUJ8BFOFOQ@VUT_Y@x@R\Ù¯D@VUTN"b\D@T$[KFHY@GXxib"vR\ªD[)WXaHR/TYRUD@[_T±xSB$F[)BZR\Q@T]HN"FOTVUYCWXTR·b h
WX[)Yb\bUV\WX]HFHYSQ@T]OxD@GXR\d F´oHFOtHY@¼ GXd FK¹J aORUT"DST>R\V[)xSWXFY@BZWXQ@Y@N"]ÀTx@YCFCR\Bb$Q@hN"¢&TZD@YWXR\GXT b$v D@TV\ T´[)V\WXTb
* ( R
J8bUFOdCV_b\RUVUTTN B·[)[KGXGbUbUR\WXD@]OVUY@TZb&bUD@WXY>FOGcxSxSTbBFKVUT·J [)tObU¼WXY@]h l·b\tHB¼FHV\hT_nOFHtHV\¼ x@ThV vc[KvXv b&N>l·tH[KtOY¼ dwhIB·RU[)D@R T p
WXbUl$5Y WXÒ FO4h Y>tHToHBZR\RU FWXFHQSGXYF)Y@[K¢WX¥GTQSb\V&RUTTB$R\N/[KTbUVUN"T$N/§ZWXv biY@]kWXYhbUR\R\D@FHTP/B¢FHGXFOGXTVUBxwR\WcVUFOTZY"N"¦8F)[qaHJ8[)RUTZGeh)V»[)PkYSTx"VrJ8BFHFOV\N"YaOWXTY@V ] p TB]OFOFHN"VUP@WXTQ@b_R\T[Kb2b4RUY@D@TT/TZx@P@TVUx±TBZQ@Wcb\YCWXRUFHWXYG»[aH[)]HGXWXQ@aHTT"Y[KV\RTB$R\D[)GX[)G=R2WXb PFO[KWXBYDSR·WcÎTZl$aHZTÏ xiv hI [KY@D@x T
VURUTZFÀbUQ@FHGXg@RUREWXY@[)WX] YlH[l bUPWXY@FO]OWXYGXTR p YCP@Q@VUTNwBZWcgkb\TWXFHVY9N"aHT·[)[)GXbUQ@QSTZVUbT"[)FKV\JÑTwb\RUdCD@bUTRUYTZNØ[·aHPkTVET[)VeJ;]OFHTV x p
x@.®.VTUaOTTAC]OGX`FHFHYÌPk^ TxÌ®47 T·gC[)dGXWcRUbD̰Â[_ACg@WXBGcWXGXg@WXW¸TGX[)Y@WXN¡ FH]OV®.[KP@TVUD@b\WXDB$[)GS[)x@Y@FCx BQ@BN"FHGXGXTZT$Y[)RM]OBQ@FHTGXb9GXTB[)R\R±WcFORUY D@HhT N>RUDS[KT>Y@[·BaHTTZFHVYw[)]OR\T"D[)P@R V\TxSBFWXbUBZWXQ@FON"Y9TaHYC[)R·GXv»Q@SFTb2HV F)[J4R\WXY@TbUx@R WXaCb\WXTx@RQF)[KJiG»x@xSFCFBBZQ@Q@N/N"TTYCYCRUR\b·b h
F)VUT'J2J8TRUVUD@TZT9Y@B^ Tb7 J;VUFHÔIN nH7 sOt x[KN/BR[)TZTb,gkx@[)WXB$`.bU[KTOY@GÊÎ qWXEaHÏ FOh&TQ@V\BbUVUFHWXYkReYS[)dCbUGXv¡WXb2bUR\J8WceVUYSR´FO]NØWXb9F)JR\[{D@o)TbUQ@dHgST$h mH[KbUÒHTVUÒbR RUQ@[)TZV\Y@N²TGXTJ8b\Q@PkbV\TRUVeFOD@J;RUFHTD@VUV&TZNVU[·¢&aH[)TZY@WXbUVBZT[)T ]ObUF)TPkx2aOTTBRUV&F_Wº¹RUFOTZDSxig@T Rh TZ[K¢YWcYwR\T"WXVU[.¢&T_]HWXb\GXGXTFHG=R$gkQ@vM[)bUeGCTY>N"×UR\T$PSD@[KVUT.bUTQ@BJ8FHWXV\b\TGXWcGXFOF)FKY@¢&JiÙ bUWXY@dCFOb]Vp h
FHRUl$DSYSOTHGcd s RUWXnHRUR\F oOGcT¯o l$h HP@HCGXSQ@m l b´v FK[KJ"j g@RUbUGXD@G&R\VTF)[)N J_BZRRUDDS[)[·T b´aHVUT[TEJ;[KTx@g@VUFCTbUYSBR\VQ@B[)TN/BZb±RUTb$YCDv R$[·v aH°9T RUT¯WXDSR\VUGcTTZTEb$J8x@TZh&FCV´g@BQSR\Q F Rp [2× jbUTZR&ª F)f J Ù.RUTRUb\FR4VUx@TEJ;FCTBV=QSRUN"F4TR\YCD@RUT b·vlHlEp PkFHWXYCR»[·aHTZV[)]OT P@V\TBWXbUWXFHY2F)aHTZV
N"¦8^«TZYTx@R\b WXB$¢[)GiTVUATQ@N>gHE[KTYCBR&Q[)®.GXGXT$d[)xSWcYSWcYSx@]HTb$ChCTx"FOVÑQS^«bUWXY@T$AC]2® bUQ@§gHWXY/ETZRUBD@R&T B$[)R\[)T]HR\WXFOFHVUYWXT[Kb G | C )6;79:$? -~Z5 ·6;:»<I
]HWXÔIY FOWXg@VURUWXVEDST[)b&T V\d,x@.®T'FK¹JY@ACT`^x^ TWXxSY>7 WcBZÌ^WcYST·x@THACvFC®BQShk N"[)D@Y@TTZx YCVUTR&lB[)FOhV\oOGcT¾GXn@Tl BZ[KRUgB$WXFO[KFHRUQ@YiTR v ]OFH°9l·V\ WcT hTZtHb&QStOP@bUt TVUx±TB$b\[)TRUR\D@YCT TRp
798;:
ÐVUTZTEN"J8FHF)V\aHT,TZx[KP@RUP@DSGXT>dCWc¢YS]9FHV\x@J8T$bw[KRUWXY,Q@V\T[±b\b\RT[)GXTYSBxRUWX[)FOV\Yx,RUb\FRUFOP,x@FC¢BQSFHN"V\xTYCGXWXRUbUb·R·h_Î l·¢ZÏ T v
TaHD@[KTGcYQk[)T$RUTZ[KxBED,¢&FKWXJ RUDÌR\D@[{T/¹YaHQST/NwJ8gT$T[KV±RUQ@F)V\J"TxSbUTWÓ³IGXTTBV\R\TWXYCFHRYRUN"TZVUTN R\D@p FCVUTZx@N"b2F)¢ÑaH[K[)b G
N"Q@l$OY@TZHWXY¥Ct R\Q@b x@T2FC[KRUBb TQSV\R\N"N"D@TwTb.YCRUWcRUTZY±b_bUR_RU[KDSb b\TT[wR_R\VR\WX[)VY [KWXY@WcRUYSWXDSY@WcWcYS]/b]"b\bURUTbUQ@TZR·x@R!v dC[K v Y@D@x TDSR\[·TD@aOVUT TT"Vl·[K[KH]HVUST T2l sHGXTx@n Y@FCh ]HtHBR\sHQ D Ò p
,
RU[)DSGXG=VUTRUb\DSD@TFHGXR\x@TVUb·N/v b j WXY´R>[9[x@DSFCWc]OBDQ@N/RUD@TYCV\TR!bUD@[KFOVUT2GXxigh4TZWXGcR/F)¢Wcb/R\PD@TwFObURUb\DSWcgSVUGcTTÀb\D@RUFHDGX[KxiR v
F)]HJ"FOVU[WXTb_x@FC[)V\BT"QSN"[)b\TbUYCWX]HRY@WXTZb x´l$R\ÒOF>s T·[)¢BEFHD9V\x@x@b$FCv B[email protected]/TYC[·R$aHv2TeVEY´[)]Ob\T FHN/l$n T_bUB$T[)YSR\bUT T p [)xSFx@[·TaOxÂFHWX[ xwN/VUTZTN"R[«F)aCVUWXQ@Y@GX]_T[)RUGXF±GIRURUD@D@TTR\P@TVUV\N/FCBb TbUJ8V\b·FHv N² [ D[Kx@RFCBWXQ@b·hÑN/[KTP@YCP@R$GXhCd9¢[T
RUgkDSTT B$[).®QSbUT{A` RUD@^ T7 x [KRB[FOVU[)P@V\QSTb N"WXb.FHN"V\TFHV\×\TY@x@FHWºWXË"bUdCBÙkQ@v GXR4R\ DD[)[KY R9 WXb$Th2Q@R\RUTD@VUTb RUTDSN/VUP@Tb\ReD@dwFHx@GXxFCBRUQ@FN/[±TYCx@R FC@BFOQSRUD@N"TZTVUYC¢&R2WXbUFHTHY@hIGX[Kd9P@P@WºJ GXd"WXR2R\D@VUTT_b\Q@BGXGXFHRUb"b\TbUWXYÂRR\[ÀD@VUY@TFHb\DY pp
¢.®FHVUACx `)B$^ [KRU7 T]O2FHvVUdLFOBY@FHbUV\TZVU¥TQSb\PTYCFORUY@GXx@dThCY@R\BZD@TTbB$[)[)R\V\TT´]HFON/VUWXFH$VU[KTRUWXFH×eJ8Y/Q@ZWcbdCN"ÙFOWXVUY T FHGXx/¢&D@WXBDV\TbUQ@GXR\b4WXY>[2Y@FOY p TN"P@RedwxSFBZQ@N"TYCR·v @*
0.85
0.1
0.8
11-pt average precision
0.75
IG
0.7 0.01
0.65
0.6 IG
DF
TS 0.001
0.55 CHI.max
MI.max
0.5
0.45 0.0001
16000 14000 12000 10000 8000 6000 4000 2000 0 1 10 100 1000 10000
Number of features (unique words) DF
Figure 2. Average precision of LLSF vs. unique word count Figure 4. Correlation between DF and CHI values of words in Reuters
0.95 1
DF-CHI plots
0.9
0.85 0.1
0.8
11-pt average precision
0.75 0.01
CHI
0.7
0.65 0.001
0.6 IG
DF
TS
0.55 CHI.max 0.0001
MI.max
0.5
0.45 1e-05
16000 14000 12000 10000 8000 6000 4000 2000 0 1 10 100 1000 10000
Number of features (unique words) DF
RUBV\FOTN"N"PTZ[KGcd±V[)[)gS]HGcT=]OVUPkTTb\VebUJ;WXFHaHVUTwNR\[)D@Y@VUBZTZTbUD@¢&FOWXGcRUxSD b_Q@UPªp RUWXF b_mOgktHT¼ R\RUTR\V·TVUv N VUA±TN/D[KF)b_aH[)[ G aHbUWX[)b\R\RUWXTFHYCY@R$b=vwFHªY WXaOTTY Q@R\RUTDSVUTwb&[)aHY@TV\x2d FHx@Y Wº³Ê.®TZVUTACYC` R^ [KP@7 P@¨GXWXB$[K[)VUR\T&WcFOD@Y WX]Hx@DSFOGcd N>B[)FHWXY@Y pb
WXY¶ 3 hk[)Y@x [)gkFHQSR ÒHtH¼ RUTZVUN£VUTZN"F)aH[)GkWXY ÔiÔiASvS°ÂWXRUD
1
Figure 5. DF-IG correlation in OHSUMED’90
DF-IG plots
D [KC¦ E b"<Z[)&§ YHÂFOGXP@FOPk] FHbUWXR\C¦ T>E P <gS§ W½[KI´b$hGXFH[)] bwB$[KG¦ Y,E§vgkT DSbUTWcbTZYÂg@W¸[)WXYb&gkRUDSTBTFOJ8N"FHV\TNwb&TZQ@ G¸[p
IG
0.01
RUTZVUN"b&B·[)YD[·aHT.D@WX]HDbUR\VUTYS]HRUDv
B$[KD@RUTTbwTCR\DB[)TGXR2GXTBYCFHRwN/PkN"TVeFOJ8YFOVURUNTV\[)N"Y@BbwT/[)F)V\JT"WXY@hx@=TTZUx,ª WXYC[KY@J;FHx,VUNL[)®.R\Wc_aOTWXY@J8x@FOW Vp
J8RUB$FOTZ[KVUg@N>R&Q@G¸B·[K[)[)RUV\WXRUd´FHTZY±]HV\FHT¢V\x@WXT$Q@VU[)BTR\R\WcWXGXFOFHFHY"Y9b\R"REWX[)R_[)bUR2¢¶Cb·FHD@v Q@WX]H;GXJ»x±Db\Y@GXWXT]HFHaOY@R_TWº¹GXgkb>B$Tw[K¦©YCTHPkvRÑFH]bU[Kv b\N"WXg@HFHGXOT2QS¼ YJ8§_R\FHb V_FKF)J¶ J=3aHWXYF pp
0.0001
1 10 100 1000 10000 100000
DF
0.4
RUWXN"Dk[)TRYCR\Bb_FHN/bUD@N"F)¢µFOY>RUDR\T[KVURN/R\b D@WX[Kb"VUTw[)bUY@b\FOQ@Y N"p WXPSYCRUJ;WXFHFHVUY±NN>[)R\[·WcaOd±THY@v FH4QRwS[)V_PSTP@CGXd´PTZR\V F p H
RUTZR&B·[)RUTZ]HFHV\WX$[)R\WcFOYiv
IG.Reuters
DF.Reuters
IG.OHSUMED
0
100 80 60 40 20 0 gkWev TTHvXhBV\WXQ@RwBx@W¸[)FCG~TJ8b/FHV.Q@bUTTÀCBB$T[)GXGXR\TTYC]HR_FOVUPdTZVeWXJ8YCFOJ8VUFHN>V\N>[KY@[KBRUTHWXFHv_Y9P@¯WV\TXbUbTZYRR[)b\¶WXYp J8V\RUTD@THT h
RULVE®4[)WX=Y@¢&WXY@D@]2WXBEbUD"TR·[)hkV\g@T&Q@RR4[KbUDk¶ [)p b4b\T[2Y@b\PkWcTR\WXVeaHJ8FOTHVUv N^[)=Y@WXBb=TRb\[KWcbUN/¶ pWXG½b\[KTV Y@b\RUWcFwR\WXUaHªÌTHhC[Kg@Y@QSx R
Percentage of features (unique words)
iÅ
Î l'Ï G¸L[)vY@]Oj QP@[)R\]OTHT&h_WXY@vx@&T PkT[KY@N"x@TTZVEY[)R4Qi[Kh.Q@[)RUY@FON>x¨[KA¤RUv&Txw°9GXT$T[KWXbUVUb·Y@v WXY@]_ F)FK¢&Ji[)RUTV\Cx@Rb
Æ Å È4CÉUÆ Å H
B$[KRUT]O FHVU WX$[KRUWXFHYN/FxSTGXb$vIeY $h l$H K v
!"$#
(:
&(:
)&H#
x@WXD@N"WXb_TZY@WXb_bUWXFH[)YkY9[)TGXWXaHRe[Kd_GXQVUT[)xSR\Q@WcFOBY´RUWXFOFKYJJ;J8T$FH[)V&R\Q@R\TVUCTwR&b\B$T[)GXTR\BTRU]HWXFOFOVUYWX$N"[KRUTZWXFHRUD@Y FC[)x@R4b[KWXGXY G Î nqÏ [)b\TbUY@FCY@BW¸TZ[)RUR\D WXFH°Y9[KY@VUFOx"VUN"LDCb·hiQ@VUNwBED QSRU[)QYS[Kx G»WXf YC[)J;FHR\VUVUWXNBE¶w[)R\® WcFO[)Y YS¶[)b·Y@vix9°9GXTFOCVUW x p
&%
U¢&ªWXRUD@[KFOY@Q@x RLGXFH®4bU WXY@N/]FHB$b\[KRRUTET³I]OFHTBV\WcR\·WXaH[)T_RUWXFOWXY>Y¾[K[K]HB]OBVUQ@TVEb\[)bUBWXaHd9TWXRUY±TZVUFHN Q@VVUTTN/CPkF)TaHVU[)W p G Î oqÏ °ÂbUTZY@WXGXbUGXWXW½RU[KWXN²aOT° GXT$[KvÊVULY@WXFOY@D@]wTY9N/[)TRUY@FCx x@bÖ J8FOFOVV&[)N RUTCARWcYSB$]H[KTRUV·Tv]OFHLVUWXFO$YC[KRURUTWXFHCYR p v
N"BFOTZN"YPR\b[KV¢&[)WXgSR\GcD"TM¶ RUF4\R\D@T&[KY@PkxTVeJ8ÔiFOÔiVUNASv[)@Y@BT±RF)Jk\D@Uª¨VUTb\[KD@Y@FHxGXx@LWXY@®4]wWX¢&bMWXJ8RUFHD2QSY@Q@xP eY
/0#21 35456" (:7 :& #8 93 H :%
ch l$OHÒ v
@( ,;/0#<`&B@()@V # B (:
RURUFF mOHt·tOpU¼ ÒHtHR\¼ TVUNR\TVUV\NÌTN"V\TF)N"aH[KF)GeaHh)[K¢&GÁD@vIWX^GXT QS RUQAw[KWXG@b WXYCBFOJ8FHN"V\N>P[K[KVRU[)WXgSFHY.GcTD¢&[)WXb=RUD/WXYCQ@J8T P p oHtOs·pUo@l·m v ;
=@! , &( $ '# @( !
VURUWXF FOV4[Pkg@TW¸Ve[)J8b.FOVUJÁN[·aH[)FOY@VUBWXY@T]wBFOVE[)N"VUT2P[KRUVUTZTVUxN"R\b_F"[KRUY@DSx¾T[FORUbUD@R\TVUFHV.YSN"]>Tb\R\TD@Y@FCb\x@WcR\b&WXax@WXQ@Red T Î KÏ!° v [)®2GXR\v@HLv VU TTVEBZ[)dx@hSWXY@Ð ]wv ^9N/v@WXP@^ bÑ[K[KbY@[Kx Y@xiN/h@TAIN"v FOv@VUAd N"J8FHWXRUV&Di¶ChOY@[)F)Y@¢&xGXTx@v ]OÔMT v
RUFwPSVUFHgk[)g@WXGXWXRedTb\RUWXN>[KRUWXFHYwTZVUVUFOVUb·v TY@Y@T]OBZRUWXY@WXFHTY TV\NWcYS[)] BED@BWXG¸Y@[)THbUvb\WºJ8dCWcYS]BTZY@bUQ@bM±VUTh R\oOQ@m VU Y@b= FOÒHY_o RUh D@l$TMOBHFHn Y v p
°9RUTZVUTwN x@WX[)bUBZV\T´F)aHb\TRUV\VUTFOxY@]HRUDGXdÂ[KRBZFHR\D@VUV\T>TG¸[)R\hTxiih_Uª£V\TaH[KT·Y@[)xGXWXY@L],®4 [¯bUP@BFOV\TVUaCTbwWcFOQ@FKJbUGXd[ >` 8? .-
Q@WXYÂY@¶CRUY@TZF)¢&RY2B$[KJeRU[)TB]OR&FH[)V\gkWc·FH[)Q@RURWXFORUYiDSv T4WX N"D@PkWXbFHV\bURQ@[)YS]O]HBT&Tb\F)RUJib BR\FODN"[)N/RFHRUDkY_[)R\R>TVUN/b Î mqÏ AIÔivi[KY@.xT[KTZQ@VU¢TV$ThIbUR\[KTY@V$x hAI v vS ® vi[K.VUQ@bUDSN>N>[K[)Wcb·Yhv ªweY@v °x@TZvISQWXY@@]/VUYkgC[)d"b$h G¸[) R\Tv YCR v %
RUaHTZTZY@VUdwbUWXaHG¸[)T V\Gc]HT·T[)RUVUTYSCWcYSR]wB$[)N"R\TT]HR\FOD@VUFCWXx@$[)b$hCR\WXb\FHQ@YwBED¾P@V\[KFHb&g@GXY@TTN/Q@VEb»[)¢&GiDSYSWcTBEReD¢[)FOVUVUT&¶Cb$FOhkRUR\D F p Î q[Ï Y@TvV·@Qh @[)DSY@V$x hiAIv¤v ® [)V\TRUVEN[)b$[)v Y@Yj [SWXV>hIKªwvIp ÔI[«Q@b\VURUQ@WX]GXT hip ^g[)vÊb\TAx9BDCNw¢&Q@[)YCGXRUR W pp
TV\¢&Wcb\T_WXYCRUV[KBR[Kg@GXTHv bUÒHREtO[)Ò·]OpUT ÒHnOWXY@o xShOTTx@CWcWXYSRUFO]_V$bUh dCbUR\TN"bMJ8FOV G¸[)VU]OT4b\Q@gHETBRK¹kh Tl$GXOx@@b$l vv eY
%
e)[ Y R\WXFHY[)YSx9Y@T¢ÌG¸[)VU ]OTwRUTCR2BFHGXGXTBR\WXFHY±J8FOV_VUTb\T$[)V\BEDiv Î Hn nqÏ XW x@v °TYCRUv.Wº¹°ÂB·[)WXGXRUg@WXFOQ@Y_V F)[)JYSb\x RUFOP¢vFHACV\WXx@VUFOb$vRU¶CWXYiv D@T,©h [Kl$Q@ RUFOCN>m @[)mHR\mWXB h
? M= 0# 1 3!
%
H
# (;!"$# `&B@()@E # B%
(:
KhkP[K]HTb l·Hn nHtSl h l·HK v
C=Y!@ V@K '&(
$ '# @( !
$l OHn v :@ ?& ? ^@ ? -
Î l$tqÏ bU2TZGcv TZBRUFOWXGXFHGcTZYV»v&[)eYSY x^9v)A@[)Dk[)N" Wev F)¢Ñ[KVUxFHPSRUWXN>[KG·J8T$[)R\Q@VUT Î nHoqÏ4ÖGXT$[)v \V Ö Y@[KWXY@Y@]]v J8V\7 FHNÌCPDCTZVUQ@RN>Y@[KTY_Re¢x@FHTV\BZ¶ Wcb\ WXFH7 Y@³IbTWXBYwR\WXRUaHTTwCR[)Y@B$[)xR\TTE]HË"FOBVUWXWXT$YC[ Rp
!"0# 1 3 .-
%
$
(:
`&(:Y
,:&Y F:#
( @#$(%
C)- ( # Oh l$HOÒ v
RUWXFOY K[ Y@xVUTR\VUWXTaH[KGÁv eY H
:%
Oh l·HHm v
,:&V# F` & $( `%
Î v )[ Y@]kv FHWXbUTV\Tx@Q@BR\WXFHYÌWXY[Âb\R[)R\WXbURUWXB$[KG" [)P p
Î l$nqÏ L&[)[·GXaCG¸Wc[)x{Yh[)vYSx ÔIT ¢&FHWXb·Y h f [)FHPSgk¶HT[@VUv R 7 VEv&[)WXACY@BEWXDY@]"[KP@[KWXVUGc]OTHFHh V\WcR\[KD@N"N"TbMb J8FOf V v n)KÏ4ÖP@V\F[)Ö BE¯D RUF R\TCR_B$[)R\T]HFOVUWX$[KRUWXFHYiveY
(: $
&(:, #C
) (B
eY
B .-
A ()@ R")& : F ( H "9
& R Î nHÒqÏ4ÖRUF¿v Ö R\T[)CYSRÍ]v B·[)j RUTZY]HFHTV\aHWX[K$GX[)QR\[)WcFOR\YiWcFOv YFKJ eY bURE[)R\Wcb\RUWXB$[KG [) PSP@VUFC [)BED
'=@ , H
M= 0# 1 3
Hh l·HK v
&( $ '# @
( ! F^ # +# (
¦¸R\F"[)P@PkT$[KV§h l$OHs v
D) >L(#K$H% ) $ 4 D ( : F"L" -
Î nHqÏ4ÖRUF.v VUÖ TN/[)Y@F)]aH[)T Y@V\x2Tx@°Q@YSv x[)vKYC°ÂR=WX¢GcgSFOQ@VUV$x@vOb`4WXY_b\WcRUYST] CR=BZFHB$VU[KPSRUQ@T]Ob FHbUVUREWX[)$R\[KWcRUb\WXRUFHWXBYb v
( # !@(
Î l$sqÏ v v 4QSWcYSG½[KYiv±Oh eY@l ¦x@l Q@Q§ BZ @RUWXl FHÊYl·tHF)Ò J4h xSl$THBOWXÒ bUWXv FHYR\VUTTb·v
C)- (# E-
%
Î l$qÏ ®BFOv.N"ACPBE[KD VUQ@WXbURUFOZTHY"hwF)J Bv j G¸[KvbUbU®.Wº¹Q@TZGXVUGeb_h [)[KY@Y@xÀx x@FCv B2vQ@N"f TZTYx@R&TV\V\TbUP@TYV\Tv bUTY j p
W
eY
()@ E:&E - (BH "9 ( R