クラスタ指向インデクシングに関する一検討

全文

(1)社団法人情報処理学会研究報告 IPSJ SIG Technical Report.

(2)

(3)

(4)

(5) . 2004−NL−159 (7) 2004／1／13. "! #%$&'()*"+ ([email protected]). ,"-/."02143"5/687"92:<;8=/>2?A@CB"D/EGF8H"I/JLKNM814O"PNQ2M8R8S2T2:/; &"'"U2V/W ?L= 68a"5/R UGb :/;8ced"f2g ."0"hLi14j2k"l"m 729"n j"k2o nrX"Y86tZ"sv[2ux\"wy]"Bv^"Dx_"Ev` z|{?v}~FrMtR1%vv:x` ;yv k ]rr Ir Y Y &|' Q%v] yrR%B|Y8Drp"q z {"?"}"EL :/;8ct<6"1B2D/z"{2?"}"E8"<"Na B"DQ /M8R8/L/8z2?"E _ :/;Lc U"_ V/ 6208B"D/E8j2Q:/;L" Y UGV 68¡/Y¢"RL/L\/LLzG?"6/£8; U"V EL¤ 5/1e¥8¦/E8§2¨NML©8ª2«/E8¬"":<;8ce®/";8¯"°<6Ls"uN±²B"D/z"{"?2}N~"F Y ª"«/E"14³ $ M8©8/L/L´Lz"?µQ5A¶·/.L¸"¡<"6L¹"º":/;/»2QL."14¼"½2¾ Y "; UGV Y8¿"À<Á8Â"Ã . ;x»Q%E%Äj/ ¿Å 6x£xÆÇ:tc An Approach to Cluster-based Indexing Akiko AIZAWA National Institute of Informatics This paper introduces a framework and implementation of an information retrieval system that utilizes clusters of similar documents. The proposed method first generates document clusters together with their representative terms and phrases based on the term distribution or term sequence match. Next, considering each document cluster as a single virtual document, an extended index is created. Upon a query submission, the system uses both the original and the extended indices and returns the integrated result. An example is shown where indices generated based on different viewpoints are used to enhance the flexibility of the retrieval system.. È ÉÊË ,-x.01%=/>?Ì@ &'UV 68ÍxÎ; ÏÐz{?"} Ñ<Ò <<z W 2FLÓxÔ cluster-based indexingÕ Y2Ö Jr×ØIrÙ6|Ú%:x; Ub ª«xE 'Û :x;%c,|-x.%dvf :x;zv{?} ÑxÒ x%rz W vF Y s,xv%Üx¢tÝ 0v1"rÞyßxà%átâ5vªãarwEyäaBvDxFtHIxJvE Urå My1"FyHIæJ%ç Y%èérY nv¨5 Áyê 5 å À kQ Qìë6îí2ïð´ñîòì< Y BGDµzì{2?ì}GE ^2_ 1t» ó E &v' OPQMtR%ôõM%R UvV 6yö $ Rr£x¶%Q45 ¶%ë Y ."; [1]c =x>?Ì@÷BD Y UV 6%Í5xR18BDxø Y 79 Á8ù ú g Á ßÆQG;»ìQL0ì1üûìý<6îþ´± Y Bìÿµ6ÍG5 R Ñ ñ ó Rw2© [2]cµ©xQ%¢ 1 Rijsbergen Ô 1979. 1. QvMyR Ï 3|5r6 Õªvã|ar0 w “theÔ clustering hypothesis” Á âv5vB|Dr0rà U|V associationÕ 6 M%R ê 5xÚ nxEy: Ò Á ;yÓNQ

(6) rR%5 ; [3] c »»y.|1/» Y ¨ Y BvD æ¶yM Y Ï ªãva wyÓ Q%0 r6v0%BD Y å À k x}yIx Ô :x BGD <z@ H Õ Y 729ì¾<6sGu´±2ë Y .G´Æ 1 U V k/6 ":<;8BGD Y ¨"¾ "¹´!Q "<6 #$´ñ ó ;8c :xv %7|9BD Y zv{?}~ØFrQ ¨BD %¨ ; ,'-/.;%c 'Y &( 0%35x!6 )*x6'Ú +:x » Y »Q%ßx/Þ .§/6 &'UV 6%Í/Î;%BDxz{ ?v}~ØvF01x M x M UvV 0k xzÌ@ H Y 2 _ 143 c 6 5 Y ,'- Q xEv%: ë Y QvM%R )|* ñ ó Rwv©%7 Y ". 8:9L<ë ;Q">ë =@± ?Þ ó R85<; Y 021 Rocchio Þe6æ£y; ¨ A C I BED Lz Ô relevance feedbackÕ . −39− 1.

(7) ."; c ¨ A I2B4D´8z0"1T"¦ Á U"V ª « Y vB Dx[4]E ¨ ¨r6F%HvIxJylxÎxMy©%ª«æ6 svuxwv1 UvV k0xz @÷HrE 3 5 :x;yÝ ` .v;%c ¨ A I B DN%zv06 YrY þ± Y &'vUV 6Úy: ; )v* Y v . ñ ó RyÍ Æ 1

(8) x.v0 Takano Þ Ô 2001 Õ Á 1 UvV 6t¨ q :r;ykQ ¨v:r;tB DxE vT¦x6 6 ñØr Á Þ UV E xá|; ê T #$x E ¿À M%R%5x; [5]c Y g `x Á Y BDxE "T¦x6%d:x; Y 6 M%Rv18BvDxz|{?v}E *vT¦x6yd|:x;%Ý ` ë4dfñ ó Re5æ;tc Cutting Þ Ô 1992 Õ Y Scat01 UV ª«B"D Y z{"?} ~F ter/Gather Qî1 ìTG¦ Y g6<£î; ¨<zì{ì?2} Y µE !G3µ#6 " %Æ $´Mî Á Þ UGV E µáì; &ì({ '#ììF Y ©Gá Y µ } ) I/?G.""; [6]c|h"©21 Zamir&Etzioni Ô 1998 Õ 0v'1 * *+, ^ - E vT M%R UV ª« Y z{v? }~FE ê 6t¤ ¶ g ` Eìd|fMì1 Web BD Y &" { '."F"68Í/Î; 0ù / ¾xE8NMLR%5/; [7]c 132 Þ Ô 1999 Õ 01 4 8BD/z{"?2} ~F"E ¨ A I B4D %zxQ [|\ ¨r%r61 5vr6 UV kr E%¤Ì¶ÇÝ ` E%df M%R%5x; [8]c p Ýx1 T¦ Á 78 :x;x»Q8x± 6 UV k z @ H Y 3 5 Eì¤ ¶ Ý ` ë4dfñ ó Re5æ;ìc AtÔ #Y 9 + A I B Drz Ô local tar&Frankel 1977 Õ v 0 1 r6v{vv>||Frñ ó © UvV ªv«|B feedbackÕ D . Y :r'Ú +æ6%surw%k r¶%M Y 79v¾x'E #0$ 7 My1/» ó E TM%R UV k xz @÷H Y 3 5 E%¤Ì¶ Ý ` ."; [9]cµh2©1îB"D %"0¨ ;0<x6LÍ/Î;80 :x!Ú + ß/Þ k /¶LM Y 7"9Gn/E #$G:/.; =?>/LÝ ` Q"MLR21 Ô Y AÏ @B Ô conceptÕ Óx6tsvu Qiu&Frei 1993 Õ ± UGV kGµ Á 2; [11]cµñGÞL621 Xu&Croft Ô 1996 Õ C . Y D 9 + l E Ô analysisÕ 021 BGD %G?¨ ;?<ß<FÞ =?automatic >µ2 GÏ @ local B2Ó µ ¶LM Y 7G9 nr'E #$:æ;xQvQëy618jvk Y ßv Æ46 HÏ @BvÓ/E Tv5x© 9 + A C I B DN%zv. UvV k xz @÷HæE 3 5 :/;Ý ` E8TG5<R85µ; [10]c p Ýµ1 I0J´Þ Ô 1999 Õ 0 =>rt7kx Ú +xEyT5xRv1 UvV k xz @÷Hx.v0. r±18Bv0D rzÌ@ HxE 3 5 :æ ; KvHxEydvf MtRy5 ; [12]c »»%.1x» ó Þ Y LM g `xY þ ±%6% M%R%5/; Y 0v1"hyi1%BvD ] k Y FyH|IxJ Á UV Y û Næ. W OI 4 6y _ ñ ó ;y°x.v;%c UvV kr%T Y P {?E UV QB|D ;<rßxÞ RS å :x; ¨xE T wv1"zv{v?}E U%r6 W ?%= X 6%ôvõ:æ;x»Qy0 v¹ MyR%5æ%5cxhv©1 L M g ` 6%ÍrÎ; UV Y V. W%v0 (rY X. BD Y Y v¸xÎv.v Æ 1 UvV ªv« BvD Y z{v?}~ØFv01Tv¦ Y ¨¹xE Z[ :x;v©á Y &{P' vF%g\QM%RyT5Þ ó ; ¨ Á þ|5c2í|ïr]^ Y zv{?}vE pX_ ¾£±Ødv: ;r»|Q Á ` d Qy;©vá1z{v?}|ø. Y' ¨vn Y Y x0h /Æ ,!-x6xÞ%%5c » ó '6 M%R%,-x."01xÞ%ßxà8á S å M%©%BD zv{|?v}vEyvx|yBvDQvMt R a5r1%ryæy z?rQM%Rtôõ MyR UvV 6 vT:x; W ?t= XYy^_ E U|b :r;%c"BvDrz{v?}vE%æ%%z|?y:x;r»Q Y ° QM%R%/E b;x»Q Á .w";%c 7G92B2D .Á c ^ ?d?8 :<; ¨<621"» ó ÞE2h • Q8áG;/»2QL6/£ ;GR"1 eìY B2D<E8j2NQ :/; ¨£<ÆëLFH2I<JµQ2MR µ6 Y 2¸<Î:<; Q Á ."w;%c FH2I<JK<6µ£<Æ ¨2n .Y Y Á Á ; / «<6 • f W²:<;<QL1ØG¹ Y l g 0 ] h -<06 i¹M8RB2D zG{2?2}2EL _ MLR21 U2V 6¸2¡/G6L0l g Ñ<Ò ¾xE%ä/©%x;x»Q Á .w;Lc 1 lonx.%BD/z{?} Y ¹ºN'Q 6 ó E%T5x© jk m c prnx.v01%j UvVæW ?%= XYy^_ 6%a5r'R

(9) 0x;yq klvm QØj|ko pvq 6 f WMym© lÇa Y B|Dxz{v?} _|` 6ya5x R sPUrM%1%rP6 tDnx.%r% %zv? Y u \ #$Q§¨ ` E%|:t c v m6 wrnr.%Ä|jr ¿ ¤ Å E x 7 M%m1 yznx.hxQ%á'E

(10) x;8c {}|}~}} 2.1 P 0P UGV 0NQ Q";8BGD Y %"¨µE 2. 1 66/h ó ;8k Y %|¨xE T Q|MyR1rÏ BvDxz{|?}yÓ/Ev18BvD%¨ Ô ∈ DÕ Q"» ó E62¬2:/;8k%"¨ S Ô ∈ T Õ Y8[ S \ ¨æ%x.y¹ºv:x;%cBvDxz{v?} Y

(11) · Y%Å E 1 6%":ecz{?"}6x6 S 1 S 6%¡x¢"R2.1 ¬T Y =x>"?Ì@ &' E%¹º M%R";%c D D. D. T. T. P6. D. 2U VxW ?%= X 68ÍxÎ";. Y ó E0 2 Q 3 6 2:ecs2,µ2Lg Y 0G1 (1) B2Dµz2{2?G} Y _ 1 (2) x% %z? Y _ 1 (3) U2V QØx% %z2?%§¨æ1 Y pÇax.;%c jPk 1 6%a5xR8Äjx6'

(12) x;%c BDxz{?"} Y _ (1) ,2-x.0%BDxz{?"}xQM%R1 hxQ%h Æ Y ";í"ï<8ð´ñ8Lj2 Y F8HGI/J2EL2¹NMR85<;8c 2.2. −40− 2.

(13) <CLUSTER> </SUMMARY> <SUMMARY> <DOCSIZE>m</DOCSIZE> <DOCLIST> ID1</ID><TEXT> 1</TEXT></DOC> <DOC><ID> ID2</ID><TEXT> 2</TEXT></DOC> <DOC><ID> ··· <DOC><ID> IDm</ID><TEXT> m</TEXT></DOC> <DOCLIST> <TERMSIZE>n</TERMSIZE> <TERMLIST> <DOC><ID> ID1</ID><TEXT> 1</TEXT></DOC> <DOC><ID> ID2</ID><TEXT> 2</TEXT></DOC> ··· <DOC><ID> IDn</ID><TEXT> n</TEXT></DOC> </TERMLIST> </CLUSTER>.

(14) . . . . . . . . B Dxz{?} Y

(15) · Å » Y £r¶Ø|z{?v}Ey _ :r;%Ý ` QMtR1©xQy¢ Dhillon Þ Ô 2001 Õ 0ì1 <ì6BìD<EìzG{G?G} ~"F<MLR6 ß<Þ S å ML.© "¬ /z @CH/E Ï @0BGÓ Q²:.; @ BGl Ô concept decompositionsÕ ` Ed2f D ÷kv¤|o Y Vv_ l| l Er6ysvu± MyRyÍ Æ [13] 1%Bv Ô 6ß";%g ` LSI Latent Semantic IndexingÕ [14] QvM%R f Wñ ó ;yc/hv ©

(16) r.0|1 Slonim&Tishby Ô 2000 Õ Y Information Bottleneck ` 1 Dhillon Þ Ô 2003 Õ Y &"' !x co-clustering " 18BD Qk Y 3 &!' # 6%s|u ± 3xz{v?}~ØF ` ëØd|f ñ ó RL5<;8ct, W ?8= X ."02140 n<!.

(17) /;/£/¶6G1 &!' # 6%su±ØkQØBD Y 3rvz{?v} ~Ø|F 1%Í£Øãæ1%j2ko pq 6%su ±ØBDxz{?} ~Ø [1] F [15] E T:x;8cN» ó ÞL061 =!$&%xß%!a '/%I } ^ - ç Y 4 E & My© 9 + yBDxzv{?}~ F ` .";%c . 1:. :/;8/L8z"? Ô jk Ï /8/8´Lz"?LÓ Õ E6 ó)(tó _ :r;%cxyxyN%zv?|01%BvDxzv{? } * Y ¬ryk ] k+,xEv1"zv{?|}E ^_ :r;%B + Dr6 4 ¸rÎ;rë Y .;ycN»v»t.1 À|¿ W ?y= X þ ^ Y BGD .Á d 8 :µ;2©2áG1OGP Y ø2 1 6G0L? øv|/ .!0 Y ëvQ%./1 1+2xyBDæz{v?} ~F Ey¤ ¶%»Qy0 Àv¿ v.0y5c » Y ©áv18+k 3 ] B D)4t6= 5x~E iy¹ My© .yxtxyN%zv?E _ :x;x»Q /Á 7 6;%c UV QØx8 %z?%§¨ (3) `

(18) Y £x¶Ø61%,"-x6%ÍxÎ;%x%/% %z? Y Wy|01L7|9vBDrEF%HvIxJyK:æ;x»Qy.1x£rÆ ê 5 ¨n Y xE é ¢x£r¶%Q5Ì¶%ë Y .v;%csv, y %z?æQØr%r% %zv?Ey³ $ 6 ^v_ :r;x» Qy.|.1 |T¦r6x£y; Uyxv%æ%%z|? Y Á ÂvÃ 6Æ 1 UV 6y¸¡æv6% l g ÑrÒ ¾xEtär©%r; »Q Á .w"; Ô 3Õ cGx8x8N8z?2'6 ":x;8k Y u \ ¸xÎrQØr% %z?y§¨r6%a|5xRC0 tnr'.

(19) x;%c. . 3:. U"V QØx% %z?%§"¨. 8:9<;<=|?>?@BA<C. BìD<zG{2?G}2E S å :µ;2©2á Y2Ö J<×²IµÙ<QGMLR21 j"kl"m Y 729n/6%s"uN±%z"{?2}N~Ø"F41îÍ £ (i) ã/1 (ii) j"k2o Y8p"q 6Ls"uN±8z"{2?2}N~2F41 Y l ax6%a5x'R

(20) /;%c 3. BDxz{?} Y _ x%N%z? Y _ 3.1 D!E!F&G!H!IKJ/LKMKNPORQ (2) BDxE%j QØ:/;% Y UV T"x% %z? Ô j rä ©|j2%k5lk m Y Y%%vp¨ q 6%Q \ s"uxMy±ØÝR`18jv.k|0l18mBDxÁ/T E rlY!S E6 k Ï s","<8z"?8Ó Õ Q1B"D/z"{"?2}"E8j"NQ . 2:. 3 −41−.

(21) Before agglomeration freq(ti, dj)=1 |ST|=3. SD. After agglomeration. freq(ST, SD)=7. 1. 1 1 0 1. 1 1. A representation of a cluster that is composed of subsets of documents and terms. Documents. 1 0. |SD|=3. |ST|=3 Documents. |SD|=3. c(ST,SD). ST. 7. Terms. Terms ST. ST. SD. SD. j"klmx6%su ±ØB"Dxz{?} ~ØF ` [1] 79:r;%BDrF%H|IxJxQ 6 ó 6 4 :x;%jvkxF%H|I 5;ec » Y u \ 6æMì© Á ;R S Y ß

(22) Þ M JvE S å :r;yc <æv6v0v1 R0 ¨ A I B D Ô M = 10Õ kxE%BDxz{?} Y *%xE 0:x;%¬ zrQØ7v9 Y[\ E%Tv5xR1 jCk Y £r¶Ø6tBDxzv{ T Y >%'I &I2BQM%R S å :x;%cxh|© UvV T Y >%I &I B620"1 S Y j2k": <R"E8T"5/;Lc|8Í/1g Y ?}E S å :x; [1]c . ":<;8kµE82¹ Y ?l g/6 i8¹2:<;<»"Q8.G1( 6 { % 6 x k E x : % ; c (1) X Æ Y ;z{?"}E% _ :x;x»Q Á .w;%c (1). y M % © x k E k v Q % M v R 1 6 E | B r D E U V ó (2) ÌÆ å : Ô S QØ:x; Õ c E Æ å : Ô S Q 3.2j2kD!o Ep)&q H!6%IKsJ/u LK±MKÝ NP` O!.Q 01%í"ïx *5jx. S å Mì©tB|D6 h ó ;tk (3) :x; Õ c k ,æE% ù :x;yBDxFyHIrJE S å :r;%c Àv¿ +, + Í £ ã 6 3 s µ E T G & ' R # 6s2u´± (4) S M8RG143"5/S62Ú Á â"5%"¨ (S , S ) E/Q/Æ YYy=æp|qx>Á%?Ì\ @ %Þ ¨ró 6;%B0D 1xÁ » ^vY þ £x± ¶d *|8 5:rjv; xÁ .1 /¿ Y - . , 6 MØBD/z{?}xQØ:x;%c =r>|?Ì@ÇEyl EM%©%ªv«x6r£%;xQt1<» ó Þ Y BvD g Y (4) .v0|1 4 6y:£r¶46v1%k QBvD Y : ¶yMt0 ê 5

(23) r . æ'Ú +r6v Æ 1 3x6 h - Á k ,r0 2 6

(24) "¤2o Y .21 S Í£ã S 6/£ ;GR82¹Nñ ó ; MtR%5r;%c/hv© 01M% R Tñ ó ;y! >x6 f WM81 (S , S ) * Y "Y kNQØBD Y : MyR ù Tr &' E 43|.%Í Æ 1©xQy¢ 576 rY Ô E 6 Kó (%ó ³ $ Y Q ¨r60 u sCU%B ][ 89] :;xY < Dxw Y £ (t , d ) t ∈ S , d ∈ S Õ Q \ 8: ¨NQ%1 (S , S ) ;</EhxQ8áR a Y ¶6|1 0 Y g Á ßÆ%QvM%Rxë ù / 6 ¶%»Q Á . Q Q \ %: ¨ Q Y 3 &! ' # Y lx'E #$ M% Á w;%c » ó 6%suxw%jko pq 6%su ±ØÝ ` .01h%i Þ vx|6%k ] BDr E %¨æßxÞ Æ Ty5rR%¤±üc » v j æ k E%jQØ:r'; ** +,rE ^_ M%Rv1 &v!' # 6x£ Y ¨ Y B|Dxz{v?} Y nr01 n Á ðñØ5 »vQ 6r£%;

(25) Y ¹ Q%1 (rY QrEhæQ%á;r» ; p|q n Á = $ j Y jkvorE ^r¢v bì1x» ó Þ Y Qy6x£y; &|' Y xY @ I B .%¹æh;y c

(26) jvk|oxE% ù :r;%BDrF%HvIxJE S å :x;%c »v»%. 6æ£y; 0v1

(27) r k Kv|Hr6yÍrÎv;% A = $æ012æÞyßxà%+á 4%= 5x~ Á ¸ é ñ ó © 546 Y =x>? @ÇE 4¿ . 6%l Ev:x;æ»Qy6x!£ ;Ry¹rá© ? 4P'%%= A FE 7;'R #$NM%R%5x;%c ! " ª« QMyR S å ñ ó ; S , S Y ! 601%B [15]cx» Y Ý ` 0Ls",µ2!6 * *?+0,µE _ :<?; >? @ Drz{v?}|'6 v:x; èéxY n¨v5x6æM%© Á ;Rv#1 " . Á ¤x¢;"©á =!$!%x%BDx6 M%R%5x;%c Á Ø6v;x£r¶Ø6 5 $KM%© $r. u \xÁ ¸rÎxÞ ó R BeDzG{ì?ì} Y S å 0Go1 µzì{ì?GA} @GQî6 ë ;vQ . 4:. T. T. ∗ D. ∗ T. ∗ D. ∗ T. T. ∗ T ∗ T. i. j. i. ∗ T. D. ∗ D. ∗ D. j. ∗ T. ∗ D ∗ D. T. D. −42− 4.

(28) ë pGq n Á8ê 5"kR,µE# Æ å M81/ßµÞ N Ô N = k/E2z"{G?"}L¬2"T Y =µ>"? @²Q2M8R S å :<;Lc 50Õ h©1 UV T Y >%I'&4I2BÇ6v0 S 6%v:x;tjk :xRvE%T|5x;%cxyÍx1%7v9n #$x6ylg ÑxÒ ¾rE är©%r;©|á61 ¿

(29) D. !"# $%&'. (*),+.-./021,35476*8:9; 4.1 <=>?@ABCDFEGDHIJK LM &NOPRQSTUVWXZY[!\]^&Z _` a bdc ê&fghSjilk mndopqr !yZ%F& stZu VZZ&',vZvVwZxFZ Z_ z]! { _|e}NO P~QSTjtf-idf % % $W^&' M Pk sZ _RfgF U Z M & [17][1][16] ' 4. ØÙ@ZÚAÛBC f gS½T¾U mZÜ s &z] _ÝÞVßR ÕFfÀgFS½ZTÀ¾Z!T¼àáâ Àã äZå ^F&¼_ Á

(30) Ææç¤ èé S½T¾ sê RëªÏÐê&z ] s & ' ã vV NO P~QST\]ì wx M mÜ íïî Q»SÀ!ðñ»ë$_fÃgFS½ÀT¾ ¿ « Õ &òZó å fg¼ô] s r õ r = 50%ö ZÝFFë»Z÷ âøúùTji Æûü ^& ' M ¯ _ýdfge÷ þÿq¸N OPÉQ»S T d¤V x FV¼ëXFÍo¼ p ÕF$%&ZZ Z_ v ªû ü Ê $ ó å â. s Ç &z]Veë_ó å ëÕ&vZ÷ M %'

(31) æÃçïó å ÷Ã_ fÀg SÀ½ïTÃ¾À ÊïËì ÁZì y T i ! ya äÀå ^¼&' Ï¼»ó å 4.2. 5. . ÐF_ZfWXOZO namazu[18] ! y~$Z_ !#" i¾N !%$. T '&ÊË! ( $ %&'*) ¸ X X P (ti , dj ) à -S $/. OÀVÀ& NTCIR1 õ10 30 § +, Tñi P (ti , dj )log I(T, D) = (1) P (ti )P (dj ) ! 4 5 ¯ 6 574 6 å ^Ã' 2 fZg ö !yZ%FWÀXF 3F t ∈T d ∈D ZF T ZZ]F_ D fZgZZ]_ P (·) ZZfg v 83V÷_ý:9%;eV <ó=Õef?> o%@=ÆÕ a CEDFCHG ¥ IJ»yÀ Kg ! ÊÀË »$Ã_Äf ! n M z]e ¡eVd&'% _f A B ' 4 ¿ V÷_x g¢Ne£fg¤dVx¥¦ §¨ M © s M geSd½T¾Ë yR $% &' %ê&ª_ N ! t ! «¬fg _ N !ªfg ¥¦ q%L ×NM%O Vß Õ fgS½Td¾!|e}N T 7& ,% WXZ ®R$_ iP (d |wi ) ≈ 1 , P (d ) ≈ 1 M &' OPÉQ»SZTFZ$_¼1þ 9P $ j i j N N ± ^ & Y e [ ! $ e % & ' Á

(32) ÆWXd ! Q%Rê&z å vF ¨ _ tf-idf ¯ pZ& idf ° (1) ±Z²¥ SUTWVXY9; 4 5 _Z]þ ) Z\[ ±³ê&ń M µ & 'Z¶· _qd¸NeOP~QS ]¼Vë_þ ¸%^%_:9%;% Z`[ : 4 V÷fdgUV LeM &½ T w idf ¹ M &' i O % % O a s ßR Õ&v s Ç 6 &' N P (dj |wi ) M ¯ = _ b fgFS½T¾Ë O Ö ^& = log (2) idfb (wi ) = log P (dj ) Ni cd %$Z÷_ ã 7 Õ eÕ f > [1][15] å $»%¼&»' º _|F}NFOP:Q»SZTVfZg¼SZ½T¾ (S , S ) ¿ WX $. M c%d % $÷: T '& Ð W T D Z^¼&_ |S !| S VRg f

(33) . w ±À^¼& idf ¹ !Á»Â À M % h R i $ % & ' D ¿ ½ÃTÀ¾i ¿ »fÃgÀ,Ã»$ P (S |w ) ≈ |S | , P (S ) ≈ D D Ä_ i N |S | !y%$ º ° M 6 j/kmlon N ¸ pV÷_fgSZ½T¾!|}NOPRQST |S | $ q:r%s%dy^& WX $. T #&. tË O. ! w u%v % P (SD |wi ) N idfe (wi ) = log = log |SN | = log (3) $_ ã 8 x u. ! y:z' $. T #& {dÐ u M &m P (SD ) Ni N M |%} $_fgS½T¾dË ¯ N 5 u 0 q¸NOªPRQST ¯ p& idf ¹ ÆÅ%'^ MÇ u 8 ú T i Ï ~ b ! : % & v V & 'fgS s ¨ È_|e} NOªPÉQSdT! ÊË^& z] d_xdÌ O ÷_: ¸ p. V y:z M%O Õ XÍe!fgÎdÏÐÑ ®Òp«¬Ó MªÔ ½Td¾Ëe. _ :i Ï NhOji % dÅ ¯ p { fg!ª{ ^Õ %'fg Ö ^&\ &eëªV÷ M ×. & Oe V &ª' _ # ] % CGRQ s y"d# : q¸dNOPRQST bc ¹ !yd%Õ $ S 6 C k » ¼ k ~ ¾ ù O Z a Z Y [ !qr~É _ × _Ó R M ¯ ^ tu M %' ÅFÉùOZ S %F!yZ%F&FvZWë Õ&'Ãó Web i. j. i. D i. D. D i. D. −43− 5.

(34) 4. 5: NTCIR. & {ð%3. (1). %S=TNV%X%Y%9%;. ªfgS½T¾ u 0 #T i´ $_¸ å pV÷ xÌ m u %m u m u !yd% s _ ò . Tji u 0 O ye!« f

(35) W stu % V &'.

(36) . [1] Akiko Aizawa 2002. A Method of Cluster-Based Indexing of Textual Data. Proc. of COLING 2002, 1-7. 2002. [2] R. K. Belew Finding Out About: A Cognitive Perspective on Search Engine Technology and the WWW Cambridge University Press. 2000. [3] C. J. van Rijsbergen. Information Retrieval Butterworth 1979. [4] R. Baeza-Yates and B. Riberio-Neto. Modern Information Retrieval Addison-Wesley. 1999. [5] A. Takano, Y. Niwa, S. Nishioka, T. Hisamitsu, M. Iwayama, and O. Imaichi. Associative Information Access using DualNAVI. Proc. of NLPRS 2001, pp. 771-772. 2001. [6] D. R. Cutting, D. R. Karger, J. O. Pedersen, and J. W. Tukey. Scatter/Gather: A Cluster-based Approach to Browsing Large Document Collections. Proc. of ACM SIGIR ’92, pp. 318-329. 1996. [7] Oren Zamir and Oren Etzioni. Web Document Clustering: A Feasibility Demonstration. Proc. of SIGIR ’98, 46–54. 1998. [8]. .0/21 5!26278* 92. :2;8!-0<"=4#>@$&?%(A'B2)C*,D+E FG2H 62)43 I , D-I, Vol. J82-D-I, No. 1, pp. 140-149. 1999.. 4. 6: NTCIR. & {ð%3. (2). ) ¸%^%_%9%;. for Computing Machinery, Vol. 24, No. 3, pp. 397417. 1977. [10] Jinxi Xu and W. Bruce Croft. Query Expansion Using Local and Global Document Analysis. Proc. of ACM SIGIR ’96, pp. 4-11. 1996. [11] Y. Qiu and H. P. Frei. Concept Based Query Expansion. Proc. of ACM SIGIR ’93, pp. 160-169. 1993. [12]. JKFL2G2\M] N . 26 7PO0QR .4ST@U )0V2W2X2Y . BCZ[ C^ , _ `;ba`c:edgf:bhi , Vol. 98, No. 58, pp.165-172. 1998.. [13] I. S. Dhillon, S. Mallela, and D. S. Modha. Concept Decompositions for Large Sparse Text Data using Clustering. Machine Learning, Vol.42, No.1, pp. 143-175. 2001. [14] S. Deerwester, S. T. Dumais, G. W. Furnas, T. K. Landauer, and R. Harshman. Indexing by Latent Semantic Analysis. Journal of American Society of Information Science, Vol. 41, pp. 391-407. 1990. [15] Akiko Aizawa Analysis of Source Identified Text Corpora: Exploring the Statistics of Reused Text and the Authorship. Proc. of ACL 2003, pp. 383390. 2003. [16] I. S. Dhillon, S. Mallela, and D. S. Modha. Information-Theoretic Co-clustering. Proc. of ACM SIGKDD 2003. pp. 89-98. 2003. [17] N. Slonim and N. Tishby. Document Clustering Using Word Clusters via the Information Bottleneck Method. Proc. of SIGIR 2000, pp. 208-215. 2000. [18] http://www.namazu.org/ [19] http://research.nii.ac.jp/ntcir/. [9] R. Atter and A. S. Frankel. Local Feedback in FullText Retrieval Systems. Journal of the Association. −44− 6.

(37)