Node Splitting Algorithms in Tree-Structured High-Dimensional ...

3 downloads 0 Views 118KB Size Report
In this paper, several node splitting algorithms for tree-structured high-dimensional indexes are studied. The algorithms are classified into three categories:.
Node Splitting Algorithms in Tree-Structured High-Dimensional Indexes for Similarity Search Yongjian Fu

Jui-Che Teng

S. R. Subramanya

Department of Computer Science University of Missouri-Rolla Rolla, MO, 65409-0350

Department of Computer Science University of Missouri-Rolla Rolla, MO, 65409-0350

Department of Computer Science University of Missouri-Rolla Rolla, MO, 65409-0350

[email protected]

[email protected]

[email protected]

ABSTRACT

 

      !#"%$#"!#'& !#&()+*,!- .  !# /01)!#$23 1$4#4 65%)7!#!# 890#:!#;!# !#" $#" !#'=?%   @ % %!# 6!#"%$#"!#)+ A '5%" / 5B0  !# % C*D)#"!#A8$E#!#GFH $2:!#& !#B C"%/# ) A5

I@$#H B#C#%+ % !## A%4A :FH $2LM8$#!H H# 5(+  !#)N  *D"%PO% /H #Q5(A&  /)OFH% $284#"%/0R$)"  # A !#+#C !#)F9#%$&% /@0;* #%9 !# /#%² Så.“zÅ Á ”E< Š< VGFWQ4!# WFH $EN %  †S# ^#2(!2 -“z !# /#  J@‰^ | ” J@:# + !# /#%LO FW9$%  E!A d& !#S*y4!# # KR y 2 $K*%H&( 4>9#%y5("%4 !# -#S#%G %G5(@FGCFW^$B!#   R < VGFGX&!2#!#-FH% $2RšU$EK# H&(!*,!#R%$ *U#%H  !# /# O!#u# y5 !2 $E % *P$#!yÅnA#%y5%  $u#!#%g ÔXô'þ>¡¥Qöd,¥)D¬¡þ[õô Qö ÷  gô(Ÿ¥.Åø

B=20 100%

Percentage of nodes accessed

95%



90%

85%

"RANDOM" "FURTHEST" "KMEAN" "SKMEAN" "SORT-MR" "SORT-MO" "MIN-R" "MIN-O"

80% 0

10%

20% Balance threshold (T)

30%

40%

œHDžŸ> T¡U£^¡4 ¡ó'¬÷%ž¡Lô%õ^ódôdþd¡ ¥.öd,¥)D¬¡ þ[õô% Nö÷% DôŸ¥QÔXø

!#" %Ü $'&( "*),+ &2Ë Î- ÒË)á. " ÉË(É)ÝBÍEßÍEË%ÎHÒá + ÉÎ6ÉO & SŠ./4ŠŠqO'= # 4# $ l /@0BO‰+™ O™"%%.q½½}< p )s m < —G!#$2B# 'O „Q< ƒ9 LO%NaC

Suggest Documents