15.1.1 Common design features of IR systems

• invsersed index
  $BE>CO(B index, $BK\$N:w0z(B
• position information
  $B@vN}$5$l$?(B inverted index $B$OC18l$NM-L5$NB>$K(B $B0LCV>pJs$bJ];}$9$k(B $\rightarrow$ phrases $B8!:w$,2DG=(B
  • ``car insurance'' $B$H$$$&(B query $B$KBP$7$F(B, $B8l$,$3$N(B $B=g$GJB$s$G=P8=$9$k(B document $B$[$&$,(B,$B$h$j(B relevant
  • $BC1$J$k0LCV$@$1$@$H(B ``car insurance rates'' $B$H(B ``rates for car insurance'' $B$r(B $BF10l$N$b$N$H07$($J$$(B $\rightarrow$ NLP reserch $B$NJ,Ln(B
  • $B8=:_$N(B IR system $B$NB?$/$O(B phrases $B$r(B pharses $B$H$7$F(B $B07$&$N$G$O$J$/(B, word $BC10L$G07$&(B(n-gram $BEy(B) $\rightarrow$ $B$=$l$J$j$N(B performance
  • collocations $B$NLdBj(B
• stop list
  • from $B$d(B could $B$J$I$N(B function words($B5!G=8l(B)$B$O(B, word-by-word $B$J8!:w$K$OITMW(B.
    stop list $B$H$7$FJ];}$7$F$*$-(B, inveted index $B$KEPO?$7$J$$(B. (Table 15.1 $B$K(B stop list $B$NNc(B)
  • inverted index $B$N(B size$B$rBgI}$K8:$i$9$3$H$,$G$-$k(B (Zipf $B$NK!B'(B..)
  • phrases $B8!:w$K$O(B function words $B$OI,MW(B $\rightarrow$ $BB?$/$N8!:w(B engines $B$O(Bstop list $B$r;H$o$J$$(B
• stemming
  • laughing,laugh,laughs,laughed $B$rF10l;k$7$?$$(B
  • suffix $B$NJQ2=$r5[<}$9$k(B
  • Lovins and Porter stemmers $B$,0lHLE*(B
  • $B$3$l$i$N(B stemmer $B$O(B $B8l$NI=AX$7$+8+$J$$$N$G0UL#E*$J6hJL$r(B $B9T$&$3$H$,$G$-$J$$(B
    (gallery $B$H(B gall $B$O(B $B$*$J$8(B gall- $B$K(B stemming $B$5$l$k(B)