BandNavi: Band-Member Backtracker Based on

J
our
nal
ofCommuni
cat
i
on and Comput
er
9(
2012)234—
241
…
N BandNavi
:Band.
.
M em ber
Backtr
acker
Based on M em ber
Hi
story I
nform ati
on M asat
oshi
Hamanaka and Mi
ki
t
o Yoshi
ya De
pa
r
t
me
n
t
o
fl
nt
e
l
l
i
ge
nt
I
nt
e
r
ac
t
i
o
n
Te
c
h
nol
o
gi
e
s
.Un
i
v
er
s
i
t
y o
fT
s
uk
u
ba
30
5—
8
57
3
1-
I
・
l
T
e
nn
ou
dai
T
s
u
k
u
ba
l
b
ar
a
ki
,
J
a
pan
Recei
ved:Sept
ember
05,20
1
1/Accept
ed:November
07,20 1
1/Publ
i
shed:Febr
uar
y 29,20 l
2
Abst
ract:A band.
member
.
backt
r
acki
ng appl
i
ca
t
i
on cal
l
ed Ba
nd
Navi
has
been de
vel
oped t
ha
t
enabl
es
a user
t
o di
s
cover
new songs
and bands
by t
r
aci
ng musi
ci
ns
a
who ha
ve
pl
ayed
i
n di
f
f
er
ent
ba
nds.Pr
evi
ous
si
mi
l
ar
i
t
y—
bas
ed s
ong r
ec
ommendat
i
on s
ys
t
ems
onl
y r
et
r
i
eve
si
mi
l
ar
s
ongs
,but
many r
et
r
i
eved r
esul
t
s
ar
e not
s
ongs
by mus
i
ci
ans
t
he us
er
l
i
kes
.By us
i
ng a
web-
mi
ni
ng t
echni
que,dat
a was
col
l
ect
ed on me
m ber
hi
s
t
or
y i
nf
or
mmi
on of
3000 bands
.Ba
n dNa
vi
enabl
es
a
us
er
t
o di
s
cover
ot
her
bands
i
n whi
c
h a m usi
ci
a
n ha
s
p
l
a
y
e
d.
Ba
nd
Na
v
i
wa
s
i
mpl
e
me
n
t
e
d
o
n t
he
i
Pho
ne
.a
n
d
e
x
pe
r
i
me
nt
a
l
r
e
s
u
l
t
s
s
ho
w t
h
a
t
70
p
e
r
c
e
n
t
o
f
s
u
b
j
e
c
t
s
u
nd
e
r
s
t
oo
d
ho
w t
o us
e Ba
ndNavi
wi
t
hi
n f
i
ve
mi
nut
es 70 per
cent
enj
oyed t
r
aci
ng t
he ba
nds
n d a
t
hei
r
me
mber
s’connect
i
ons,a
n d 52 per
cent
di
s
cover
ed a pr
evi
ous
l
y unknown band t
hat
t
hey l
i
ked.
Key words:M usi
c
r
ecomme
nda
t
i
on,ba
n d-
member
,web
mi
ni
ng t
echni
que,ba
n d net
wor
k,i
Phone
appl
i
cat
i
on.
I.IntrOductj0n buy songs
di
scover
ed by usi
ng t
he navi
gat
i
on f
unct
i
on.
W e can now easi
l
y l
i
st
en t
o and st
ore enor
m ous
numbers
of songs on M P3 pl
ayers
and smar
tphones.
BandN avi
di
spl
ays
bands and t
hei
r members
by usi
ng band-l
i
neup hi
st
ori
es.
W e can al
so sel
ect
and buy songs
ver
y easi
l
y t
hrough A musi
ci
an wi
l
l
of
ten have pl
ayed i
n or
wi
t
h more onl
i
ne musi
c st
or
es.However,users
m ay not
be abl
e to di
scover
many songs
t
hey may l
i
ke because t
hey spend ver
y l
i
t
tl
e ti
me choosi
ng songs
t
o buy even though t
hey spend use a l
ot
ofti
me l
i
s
teni
ng t
o musi
c.
Ther
ef
ore, we have devel
oped a band—member backt
r
acki
ng appl
i
cat
i
on cal
l
ed BandNavi
t
hat
enabl
es a user
t
o di
scover
previ
ousl
y unknow n songs
or
ar
t
i
st
s
whi
l
e l
i
s
t
eni
ng to anot
her
song or
ar
ti
st
.The mai
n advant
age of BandNavi
i
s
t
hat
i
t
has
t
he f
ol
l
owi
ng t
wo m odes.
(
1
1
Navi
ga
t
i
on
mode
In t
he navi
gat
i
on mode, a user
can di
scover
connect
ed ar
t
i
st
s
and songs whi
l
e l
i
st
eni
ng t
o a song on am usi
cpl
ayer.
(
2)Medi
a acc
es
s
mode
In the medi
a access
mode.a user
can l
i
s
ten t
o and Corre
spondi
ng author: M asa
t
os
hi
Hamanaka, Ph.
D.
pr
of
es
s
or
, res
ear
ch f
iel
ds
: mus
i
c i
nf
or
mat
i
on pr
oc
es
si
ng E-
mai
l
:
h
a
ma
na
k
a
@i
i
t
.
t
s
uk
ub
a.
ac
.
j
p.
t
han one band;he or she may have been a gues
t
musi
ci
an and/
or
have been a member
of a number
of bands.Ther
ef
ore,r
el
ati
onshi
ps bet
ween musi
ci
ans
and bands for
m a com pl
ex net
wor
k.W e cal
l
t
he net
wor
k for
m ed by connecti
ng t
he af
il
i
at
i
ons
bet
w een bands and t
hei
r
me
mbe
r
s
t
he
Band Ne
t
wor
k (
Fi
g. 1
)
.
BandNavi
can di
spl
ay t
he bands
and musi
ci
ans
t
hat
are connect
ed t
o the s
ong current
l
y bei
ng l
i
st
eni
ng to by t
raci
ng t
hi
s
net
wor
k.
To ind f
t
he net
works
of connect
i
ons
bet
ween bands and thei
r members,we use web—
mi
ni
ng t
echni
ques.
However, t
her
e are t
wo probl
ems
wi
t
h usi
ng t
he net
works
acqui
r
ed by web—
mi
ni
ng t
echni
ques
for
t
r
aci
ng m usi
ci
ans’hi
st
ori
es.The f
irs
t
pr
obl
em i
s
t
hat
web-
mi
ni
ng t
echni
ques
are not
per
f
ect
and someti
mes
col
l
ect
wr
ong i
nf
or
m at
i
on. Even i
nf
or
mat
i
on f
rom Wi
ki
pedi
a has
some err
ors,and i
t
i
s
ver
y di
i cul
f
t
t
o det
er
m i
ne t
hese erors
by usi
ng web—
mi
ni
ng t
echni
ques.
BandNavi
:Band-
Mem ber
Backtr
acker
Based on M em ber
Hi
st
or
y I
nf
or
m at
i
on 235
expl
ai
ns
t
he experi
ment
al
resul
t
s, and Sect
i
on 7 concl
udes
t
he paper.
2.Rel
at
ed W ork Previ
ous musi
c-
retr
i
eval
met
hods t
hat
use queri
es
s
uch a
s
s
i
mi
l
a
r
i
t
y—
bas
e
d[
1-
3]s
ea
r
c
hi
ng
a
nd t
ext
—
bas
ed
s
e
ar
chi
ng【
4]a
r
e
us
ef
ul
f
or
nar
r
owi
ng a
l
a
r
ge
number
of musi
cal
pi
eces,but
af
ter
the l
i
s
t
of r
anki
ngs
has
been Joe L
ynn
Tur
ner
Fi
g. 1
Net
work showi
ng connecti
ons
bet
ween m us
i
ci
ans
provi
ded, we have t
o l
i
st
en t
o songs
one—by—one because no consi
derat
i
on has been gi
ven to f
indi
ng and bands
.
songs
one f
or
whi
ch has
an af
i ni
t
y f
rom the l
i
st.Thus
The wrong i
nfor
m at
i
on obst
ructs
t
he users
t
ryi
ng t
o user
s
s
pend a l
ot
of t
i
me l
i
s
teni
ng t
o new songs whi
l
e sear
ch f
or
and f
ind ar
t
i
st
s t
hey l
i
ke. The second not
al
ways
di
scoveri
ng ones
t
hey l
i
ke.
probl
em i
s t
hat
i
t
i
s
very compl
i
cated and di
i cul
f
t
t
o Mus
i
c
r
ea
m [
5】makes
i
t
pos
s
i
bl
e
t
o
i
nt
e
r
ac
t
wi
t
h show al
l
t
he members
of a band i
n whi
ch many m any musi
c col
l
ect
i
ons
by appl
yi
ng a si
mi
l
ar
i
t
y-
based m usi
ci
ans
have pl
ayed.For
exampl
e,W hi
t
esnake has
st
i
cki
ng f
uncti
on whi
ch at
taches
si
mi
l
ar musi
cal
pi
eces had more
t
han t
hi
rt
y members
i
n tot
a1
.If t
he syst
em t
o a musi
cal
pi
ece sel
ect
ed by the user.However,
onl
y shows a l
i
st
of W hi
tesnake’
s
pas
t
and pr
esent
M usi
cream has t
he same pr
obl
em as
previ
ous
member
s.i
t
i
s
ver
y di
i cul
f
t
to di
scover cl
osel
y rel
at
ed musi
c—
r
e
t
r
i
eval
met
hods[
1
-
4]
,be
ca
us
e
us
er
s
do
not
bands
l
i
ke Deep Pur
pl
e.
al
ways
l
i
ke songs
t
hat
are si
mi
l
ar
t
o ot
her
songs t
hey To sol
ve these probl
ems.BandNavi
l
i
st
s
members
l
i
ke.
of a band f
ound whi
l
e web mi
ni
ng i
n or
der
of On
t
he
ot
he
r
hand,c
ol
l
abor
a
t
i
ve
f
il
t
e
r
i
ng [
6,7]
i
mpor
tance.Ther
efor
e, a user can di
scover ar
ti
s
t
s enabl
es us
ers
wi
t
h si
mi
l
ar
t
ast
es t
o be f
ound and cl
osel
y rel
at
ed to hi
s/
her
favori
t
e band by tr
aci
ng t
he recommends
these si
mi
l
ar
users’ favor
i
t
e songs
by connecti
ons
oft
he m or
e i
m port
ant
band members.
usi
ng t
hei
r song—
pur
chase hi
st
ori
es.However t
hese ,
I
n t
he experi
ment
, we eval
uated how wel
l
web m et
hods
have di
i cul
f
t
y r
ecommendi
ng s
ongs that
are mi
ni
ng col
l
ect
s
t
he names
of band members.W e used not
fam ous.because i
t
i
s di
艏 cul
t
t
o col
l
ect
unfamous
50 bands
for t
hi
s
exper
i
ment
,and 86 percent
of resul
t
s
bands’songs
f
r
om song—
purchase hi
st
ori
es
. consi
s
t
of t
he top f
ive m os
t
i
mpor
tant
m em bers
of W e t
hi
nk one way for
di
scoveri
ng an unknown di
f
ferent
bands.W e al
so conduct
ed an experi
ment
musi
cal
m ast
er
pi
ece i
s
t
o recei
ve recommendat
i
ons wi
t
h
1
7 s
ub
j
ec
t
s:70 per
cent
under
s
t
ood how t
o us
e
rom peopl
f
e who are knowl
edgeabl
e about
musi
c
BandNavi
wi
t
hi
n f
ive mi
nu
t
es
,70
per
c
ent
s
e
n
j
oyed
because t
hey probabl
y know t
he backgr
ound t
r
aci
ng t
he ba
n ds and t
hei
r
m embers’connecti
ons,and i
nf
or
m at
i
on of songs.Our
goal
i
s
to devel
op a syst
em 52 percent
di
scover
ed a band t
hey l
i
ked but
di
d not
t
hat
ext
ract
s
deep knowl
edge about
songs by web know befor
e.
mi
ni
ng t
hat
enabl
es
a user
t
o f
ind songs
i
n t
he same Thi
s
paper
i
s
organi
zed as
f
ol
l
ows.Secti
on 2 br
i
ef
ly di
scusses
t
he rel
at
ed wor
ks. Sect
i
on 3 gi
ves
an ,
way as
a person who i
s
knowl
edgeabl
e about
musi
c.
The
i
dea
behi
nd M us
i
cRa
i
nbow 【
8】i
s
s
i
mi
l
ar
t
o our
s
。
over
vi
ew of BandNavi
and di
scusses t
he pr
obl
ems i
n because M usi
cRai
nbow ext
r
acts
knowl
edge f
r
om the devel
opi
ng t
he syst
em and t
hei
r
sol
ut
i
ons.Secti
on 4 web.
M usi
cRai
nbow di
s
pl
ays
s
ongs
ci
rcul
arl
y expl
ai
ns
t
he user i
nt
er.f
ace of BandNavi
.Secti
on 5 dependi
ng on audi
o—based si
mi
l
ari
t
y and adds
wor
ds
descri
bes
how t
o f
ind t
he Band Net
wor
k.Secti
on 6 retr
i
eved fr
om web pages t
o descri
be each ar
ti
st
. BandNavi
:Band-
M em ber
Backtr
acker
Based on M em ber
Hi
st
or
y I
nf
ormat
i
on m em bers。 and gues
t
and/or
sessi
on musi
ci
ans
f
or
r
ecor
di
ngs and/
or l
i
ve perf
or
m ances.If we di
spl
ay t
hi
s
compl
i
cated net
wor
k on i
Phone as a cobweb,i
t
i
s ver
y di
i cul
f
tt
o fol
l
ow.
237 3.
2.
2 Aut
omati
c Acqui
si
t
i
on of Band M ember
N am es
To acqui
re names
of a band’
s member,we use web-mi
ni
ng t
echni
ques
i
n whi
ch we f
irs
t
col
l
ect
web To si
mpli
fy t
he net
wor
k,we can omi
t
t
he gues
t
and pages
about
the band by usi
ng Googl
e sear
ch API
and sessi
on musi
ci
ans
t
hat
provi
de weak connect
i
ons
t
hen acqui
r
e t
he band members’ names
by a between bands. However, om i
t
t
i
ng t
hese weak pater
n—
mat
chi
ng met
hod.
Because web—
mi
ni
ng connecti
ons i
s pr
obl
emat
i
c because the omi
tt
ed techni
ques somet
i
mes
col
l
ect
wr
ong resul
ts,we al
so i
nfor
m at
i
on can be i
mport
ant
for
t
he user
who want
s
prepared a manual
edi
t
i
ng opt
i
on f
or
band member’
s
know about
t
he band m ore
deepl
y.
names.Onl
y t
he bi
g f
ans
of a par
ti
cul
ar
band can use 3.1.
2 How t
o Acqui
r
e M em ber
s’Names
t
hi
s manual
edi
t
or,because t
hey ar
e ass
umed t
o be
Wi
ki
pedi
a and M usi
cBr
ai
nz have
pages f
or bands
knowl
edgeabl
e about
t
he band.The BandNavi
hel
ps
t
hat
l
i
s
t
t
hei
r
mem bers’names.However,t
hese pages
t
he user
to edi
t
i
ng t
he i
nf
or
m ati
on of band m em bers’
ar
e not
enough,because many new or
mi
nor
bands do names
by di
s
pl
ayi
ng t
he web pages f
rom whi
ch t
he not
have pages
on Wi
ki
pedi
a or
M usi
cBr
ai
nz,and even band m embers’names was
extr
act
ed by web—mi
ni
ng when they do,most
of t
hese pages do not
have t
he t
echni
ques.
i
nf
ormati
on about
t
hei
r
members.Fur
thermore,guest
or sessi
on musi
ci
ans are not
ment
i
oned on m ost
band 4.U ser Int
erf
ace pages on W i
ki
pedi
a and M usi
cBrai
nz,s
o i
nf
or
m ati
on BandNavi
consi
s
ts
of a cl
i
ent
appl
i
cati
on on t
he or
f
cons
tr
uct
i
ng t
he net
wor
k cannot
be suf
i ci
f
ent
l
y i
Phone and a ser
ver appl
i
cat
i
on t
hat
aut
omat
i
cal
l
y acqui
red rom f
t
hese pages.
acqui
res
t
he band m em ber
s’ nam es. The BandN avi
cl
i
ent
appl
i
cat
i
on has
t
hree modes:navi
gati
on mode 3.
2 Sol
ut
i
onsfo,The
se
Pr
obl
ems
0ur
sol
ut
i
on f
or
t
hese probl
ems l
s
to i
nt
r
oduce web (
s
e
ct
i
on 4.
1
)
,medi
a ac
ce
ss
mode(
s
ec
t
i
on 4.
2)
,and br
ows
e
mode(
s
e
ct
i
on 4-
3)(
Fi
g.3)
.
mi
ni
ng—methods t
hat
col
l
ect
t
he i
nf
or
m ati
on of m embers’name by tr
awl
i
ng t
hrough web pages.
3.
2.
1
I
nt
roduce I
mpor
t
ance Level
of M embers
To sol
ve t
he pr
obl
em of compl
i
cat
ed net
wor
ks,we i
nt
roduce
an i
mpor
tance l
evel
f
or
mem ber
s. W e 4.
1
Navi
gat
i
onM ode I
n navi
gat
i
on mode,we can t
race musi
ci
ans’wor
k by i
terat
i
vel
y l
i
st
i
ng band members
and ot
her
bands
i
n whi
ch each member
has
pl
ayed.
cal
cul
at
e t
he i
mpor
tance l
evel
of each m ember i
n a W hen a user
pl
ays
a song usi
ng t
he BandNavi
cl
i
ent
band by usi
ng document
f
requency i
n t
he web pages appli
cati
on on an i
Phone, t
he cl
i
ent
appl
i
cat
i
on when col
l
ect
i
ng band member
s’names usi
ng Googl
e aut
omat
i
cal
l
y acqui
r
es
and di
spl
ays member
s’names API.Thus
BandNavi
l
i
st
s band members
i
n t
er
ms of by communi
cat
i
ng wi
t
h t
he
dat
a
ba
s
e
on t
he
s
e
r
ver(
Fi
g.
t
hei
r
order
of i
mpor
tance.For
exampl
e,a band mem ber
3a
)
.The
n,bands
i
n whi
c
h a
mus
i
c
i
an ha
s
pl
ayed
a
r
e
who pl
ayed i
n t
he band f
or
a l
ong t
i
me and i
s of
t
en di
s
pl
a
ye
d whe
n a
us
e
r
t
aps
hi
s
/
he
r
name(
Fi
g.3b)
.
Ne
xt
,
m ent
i
oned on t
he web pages
i
s
l
i
st
ed hi
gher t
han a names
of anot
her
band’
s
member
s ar
e di
spl
ayed when band member
who i
s
i
nf
requentl
y ment
i
oned on a us
e
r
t
a
ps
t
hat
ba
nd’
s
name(
Fi
g.3
d)
.
t
he web pages.Theref
ore a user can expl
ore t
he The band—
me
mbe
r
l
i
s
t(
Fi
gs.3a
a
nd
3d)and Band Net
wor
k by consi
deri
ng t
he st
r
engt
h of bands
—
pl
ayed—
i
n l
i
s
t (
Fi
g. 3b) a
r
e
or
der
ed by rel
at
i
ons
hi
p bet
ween bands and t
hei
r past
and pr
esent
i
mpor
tance l
evel
s. In t
he band—
member
l
i
st
, each m em bers.
me
mber
’
s
name
has
t
he
r
el
e
vant
i
ns
t
r
ument
(
s
1
wr
i
t
t
en
240
BandNavi
:Band—
Mem ber
Backtr
acker Based on Mem ber Hi
stor
y I
nform ati
on of extr
act
ed web page,and t
he l
ocati
on i
n t
he web page met
al
because t
he l
i
neups
of heavy m et
al
bands of
ten where t
he pat
t
er
n was
m at
ched.
change,so f
ew bands
ar
e i
sol
ated f
rom ot
her
bands
i
n t
he Band—
Net
wor
k.
5.
3 M emberName Fi
l
t
eri
ng I
n t
he member
name f
il
teri
ng,the ser
ver appl
i
cati
on 6.
1
Eval
uat
i
o
n o
f Aut
omat
i
c
Ac
qui
si
t
i
on o
f Band M embers’
Nnm es r
emoves t
he names
of musi
ci
ans who have not
pl
ayed i
n the band f
rom musi
ci
ans’names extr
act
ed i
n secti
on 5.
2.The member
name il
f
t
er
i
ng consi
st
s oft
wo st
eps.
W e eval
uat
ed how wel
l
t
he syst
em f
o und t
he band members’ names. Thi
s
eval
uat
i
on r
equi
red us
t
o (1
)Re
move
wr
ong mus
i
ci
a
ns
name
pr
epar
e t
he corr
ect
dat
a sets
of band names
and band The syst
em r
emoves names
of m usi
ci
ans t
hat
have members’names.W e sel
ect
ed 50 bands
randoml
y fr
om ver
y l
ow possi
bi
l
i
t
y of bei
ng band members.Fi
rs
t,t
he t
he 3 54 bands
and m anual
l
y i
nvest
i
gat
ed t
he member
s’
s
ys
t
em c
al
c
ul
a
t
es
t
he
doc
ume
nt ̄e
quenc
y(
DF)
,whi
ch
names,i
ncl
udi
ng t
hose of guest
or
sessi
on musi
ci
ans,
i
ndi
cat
es
how many t
i
mes
t
he name appear
s i
n t
he by usi
ng of
i ci
al
web pages
and CD bookl
et
s.Then we col
l
ect
ed web pages.The DF i
s nor
m al
i
zed f
rom 0 to 1.
as
ked someone who knows a l
ot
about
heavy met
al
Second,the s
yst
em rem oves
t
he m usi
ci
ans’names
f
or
musi
c t
o ver
if
y t
he r
esul
t
s manual
l
y.
whi
ch
DF i
s
l
owe
r
t
ha
n t
he
t
hr
e
s
hol
d(
t
h)
.
W e eval
uat
ed t
he Recal
l
and Preci
si
on of BandNavi
.
(
2)Ext
r
ac
t
band
member
col
l
oca
t
i
on Recal
l
and Preci
si
on dif
fer
dependi
ng on t
he num ber of The s
ys
tem ext
ract
s
band member
col
l
ocat
i
on t
o ind f
m usi
ci
ans
to be i
nvest
i
gat
ed f
rom t
op D F m usi
ci
an t
he actual
band m embers,because we can ass
ume the wher
e t
he BandNavi
shows
t
he band.
member
l
i
st
t
hat
same band members
ar
e menti
oned on nearl
y ever
y i
s soaed by DF.Tabl
e 1
shows t
he r
esul
t
s
of t
he web page about
the band.
Fi
rs
t
,t
he syst
em sel
ects
t
he two musi
ci
ans
t
hat
have t
he hi
ghest
DF as
t
he
most
i
mpor
t
ant
members
of t
he band.Second,t
he syst
em sort
s
t
he t
upl
es
f
rom t
he same w eb pages
and t
hen di
vi
des them i
nt
o gr
oups
w her
e t
he pat
t
er
ns
f
or
extr
act
i
ng m usi
ci
ans’nam es
ar
e changed.
experi
ment
.Bot
h Recal
l
and Preci
si
on ar
e over
0.
7 when we use al
l
members
or
f
eval
uat
i
on.W hen we use t
he t
en member
s
wi
t
h t
he hi
ghest
DF,Preci
si
on rai
ses
to 0.
81.W hen we use t
he f
ive member
s wi
t
h t
he hi
ghest
DF。Pr
eci
si
on rai
ses even m or
e to 0.
86.
6.
2 Ev
al
u
at
i
on
b
y a Que
s
t
i
onnai
r
e
Fi
nal
l
y, t
he syst
em extr
act
s band mem bers
t
hat
col
l
ocat
e i
n a same gr
oup as
t
he two mos
t
i
mpor
tant
members.
I
n Fi
g.4,Name 1
and Name 2 i
ndi
cate t
he i
mport
ant
m embers.Name 3 i
s ext
ract
ed as
a band m ember,
because N am e 3 i
s
i
n t
he sam e group as N am es
1
and 2.
On t
he ot
her si
de,Names
4 and 5 are r
emoved,because Name 4 and 5 ar
e not
i
n t
he same group as
Names
1
w e prepar
ed quest
i
onnai
res about
BandNavi
and pos
t
ed t
hem at
t
he fol
l
owi
ng URL:
ht
t
p:
/
/
musi
cdb.
i
i
t
.
t
s
ukuba.ac.
j
p/
bandna
vi
.
ht
m1
.So
f
ar
,
we have r
ecei
ved 1
7 responses
f
rom user
s.The r
esul
t
s i
ndi
ca
t
e
t
hat
70
per
c
ent
of
s
ub
j
ec
t
s
under
s
t
ood
how t
o
use
t
he BandNavi
wi
t
hi
n f
ive mi
nut
e.70 percent
e
 ̄oye
d t
r
aci
ng t
he
bands
’a
nd t
hei
r
me
mbe
r
s
’
connect
i
ons, and 52 percent
di
scovered a new band and 2.
6.Experi
m ent
al
Resul
ts
Tabl
e
1
Recal
l
and preci
si
on f
or
acqui
si
t
i
on of
band m em ber
s’nam es
.
W e i
nput
t
o t
he syst
em the names
of 3 54 bands
cl
assi
ied as ‘
f
‘
heavy met
al
” appeari
ng on t
he home pages
of r
ecor
d compani
es,resul
t
i
ng i
n t
he syst
em col
l
ecti
ng 3,
508 musi
ci
ans’names.W e sel
ect
ed heavy BandNavi
:Band-
Mem ber
Backtr
acker
Based on M em ber
Hi
st
or
y I
nf
or
m at
i
on t
hey l
i
ke.However,f
ive per
cent
needed mor
e t
han 1
5 [
2]
F.Vi
gnol
i
.S.Pauws.A mus
i
c
r
et
r
i
eva1
s
ys
t
em bas
ed on us
er
-
dr
i
ven si
mi
l
ar
i
t
y and i
t
s
eval
uat
i
on. i
n:Pr
oc.of
mi
nutes
to underst
and how t
o use BandN avi.W e pl
an t
o i
nvest
i
gate why some peopl
e t
ook s
uch a l
ong ti
me 24l
I
SM I
R 2005.2005.PP.272.
279.
【
3】
E.Pampal
k.A MATLAB t
ool
box t
o comput
e
mus
i
c
si
mi
l
ar
i
t
y f
rom audi
o i
n:Pr
oc.of
I
SM I
R2004.2004.PP.
t
o unders
tand BandN avi
, and we w i
l
l
i
mpl
ement
254.
257.
changes
i
f necessar
y.
ng,A.
F.Sm ea
t
on.Eval
uat
i
ng a musi
c i
nf
or
ma
t
i
on 【
4】
T.Sodi
r
et
r
i
eval
s
ys
t
em — TREC s
t
yl
e.i
n:Pr
oc.of I
SM I
R2002.
7.Concl
usi
ons
W e devel
oped BandNavi
,whi
ch enabl
es
a user t
o 2002,PP.7
1
-78.
[
5】
i
nt
e
r
f
ace f
or
s
t
re
ami
ng, s
t
i
cki
ng, s
or
t
i
ng,and
r
ecal
l
i
ng sear
ch for
new songs and art
i
st
s
usi
ng an i
Phone cl
i
ent
appl
i
cat
i
on based on rel
at
i
ons
hi
ps
bet
ween bands
and mus
i
c
alpi
eces
,i
n:
Pr
oc.oflSM I
R 2005,2005,PP.
404-
411.
【
6
】
W.
W . Cohen. W . Fa
n . W eb.
col
l
abor
at
i
ve f
il
t
er
i
ng:
Recomm endi
ng mus
i
c
by cr
awl
i
ng t
he
we
b,Comput
er
band members.
The BandNavi
i
s
now avai
l
abl
e t
hr
ough Appl
e’
s
M .Got
o. T. Got
o. Musi
cr
eam :new mus
i
c pl
aybac
k Ne
t
wor
ks
33
f
2000)685.
698.
[
7J
A.Ui
t
denboger
d,R.va
n Schyndel
,A r
evi
ew of
f
act
or
s
App St
ore.I
n t
he experi
ment,we onl
y used 357 heavy af
f
ect
i
ng musi
c r
ecommender
s
ucces
s.
i
n: Pr
oc.
met
al
bands;we have si
nce added 3000 m or
e bands
I
SM I
R2002.2002.PP.204.
208.
fr
om t
he r
ock and pop genr
es.
W e pl
an to ext
end t
he syst
em so that
user
s
car
l
f
ind new musi
c by usi
ng t
he i
nf
ormati
on about
t
he names
of not
onl
y band m em ber
but
al
so com posers,ar
ranger
s,
pr
oducers,
or
r
ecordi
ng st
udi
os
acqui
r
ed by web—
mi
ni
ng t
echni
ques.
R ef
erences
[
1
】
G.Tz
a
ne
t
a
ki
s
,P.Co
o
k,Mu
s
i
c
al
ge
n
r
e
c
l
a
s
s
i
ic
f
a
t
i
on
of
audi
o si
gnal
s
,i
n:IEEE Tr
a
ns.on Speech and Audi
o Pr
oc.
,
2002,PP.293-
302.
【
8】
M.
G.
E.Pampal
k.M us
i
c
Rai
nbow:A new us
er
i
nt
er
f
ace t
o di
s
cover
ar
t
i
s
t
s
us
i
ng audi
obas
ed s
i
mi
l
 ̄i
y t
n d a
web.
bas
ed l
abel
i
ng.i
n:I
SM I
R I
nt
er
nat
i
onal
Conf
er
e
nce
on M us
i
c
I
nf
or
mat
i
on Ret
r
i
eva1
.2006.PP.367.
3
7O.
.K.I
s
hi
da,T.
Ni
s
hi
mur
a,
[
9】
Y.M atsuo.J.M ori,M .Hamasaki
H.Takeda,K.Has
i
da,M .
I
.
s
hi
z
uka
et
a1
.
.Pol
yphonet:
An advanced s
oci
al
net
wor
k ext
r
act
i
on s
ys
t
em.J
our
na1
of
W eb Sem a
nt
i
cs
5(20071
262—
278.
abl
e
onl
i
ne
at
:ht
t
p:
/
/
www.
yout
ube.
com .
【
l
O
】
YouTube,avail
Tune
M us
i
c
St
or
e.
avai
l
abl
e onl
i
ne
a
t:
【
1
1
】
Apple i
ht
t
p:
/
/
www.
appl
e.
com/
i
t
unes
.20l1.
【1
2】 Googl
e Code,avai
l
abl
e
onl
i
ne
at
:ht
t
p:
/
/
code.
googl
e.
com,
20l1
.