• No results found

Research on Collaborative Filtering Algorithm in Microblogging Recommendation

N/A
N/A
Protected

Academic year: 2020

Share "Research on Collaborative Filtering Algorithm in Microblogging Recommendation"

Copied!
11
0
0

Loading.... (view fulltext now)

Full text

(1)

) 6 1 0 2 I T I E C I ( n o it a z il a u t c e ll e t n I d n a y g o l o n h c e T n o it a m r o f n I c i n o r t c e l E n o e c n e r e f n o C l a n o it a n r e t n I 6 1 0 2

8 7 9 : N B S

I -1-60595- 43 -9 6

m

h

ti

r

o

g

l

A

g

n

i

r

e

tl

i

F

e

v

it

a

r

o

b

a

ll

o

C

n

o

h

c

r

a

e

s

e

R

n

o

it

a

d

n

e

m

m

o

c

e

R

g

n

i

g

g

o

l

b

o

r

c

i

M

n

i

h

Z

d

n

a

g

n

a

h

Z

n

ij

g

n

i

X

,

g

n

e

h

Z

i

p

o

a

i

X

,

i

L

n

u

D

i

y

u

n

Z

h

e

n

g

T C A R T S B

A

d n a m e d d e z il a n o s r e p e h

T of usersin socia lnetworksis an importan tfactor o f r

e s u e h

t saitsfaciton . Integrated user microblog conten t and user interaciton n

o it a m r o f n

i ,introduce the concep tof interes tsimliartiy and interaciton trust. This m h ti r o g l a n o it a d n e m m o c e r g n i r e tl i f e v it a r o b a ll o c d e v o r p m i n a s e s o p o r p r e p a

p .

Compuitng user interes tsimliartiy and user interaciton trust ,linear ftiitng of the 2 .

y ti r a li m i s b u

s I tcan calculate the tota lsimliartiy between the users fo micro-blog r

d e z il a n o s r e

p ecommendaiton . The experimenta l resul t shows tha t the new f o n o it c a f s it a s r e s u d n a y ti l a u q , y c a r u c c a e h t e v o r p m i y l e v it c e f f e n a c m h ti r o g l a

. s k r o w t e n l a i c o s n i m e t s y s n o it a d n e m m o c e r

N O I T C U D O R T N I

, r e tt i w T ( a i d e m l a i c o

S Microblog , c.et ) is especially rapid development ,has d i p a r e h t h ti w . a i d e m t e n r e t n I g n i g r e m e e h t n i e c i v r e s t n a t r o p m i f o d n i k a e m o c e b

o r c i m f o e s a e r c n

i -blog, to query the mos trelevant topic in the mass of conten tis t

s y s n o it a d n e m m o c e r e h t d n a , tl u c i f f i d y r e v e m o c e

b em can recommend the mos t

d e t s e r e t n

i ininformaiton,t oenhancet heusersaitsfaciton.

r e h t , k r o w t e n l a i c o s g n i g g o l b o r c i m e h t n i , t n e s e r p t

A e are three main

s y a w d n e m m o c e

r :First ,according to the registered informaiton ,choose a higher c

e r e e r g e d g n i h c t a

m ommended ,butt hisapproacht endst oproduceal o tofresutlst o

_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

L n u

D i,Xiaop iZheng ,Xingijn Zhang ,Zhiyun Zheng ,Schoo lo fInformaiton Engineeirng , ,

1 0 0 0 5 4 u o h z g n e h Z , y ti s r e v i n U u o h z g n e h

(2)

e l b a n u r e s u e h t o t d a e l , n r u t e

r tochoose;t hesecondi sbasedont hepopulartiyoft he t o h a d n e m m o c e r , c i p o

t microbloggingtopic ,bu tahott opicforal lmicro-blogusers , e c n e u l f n i e h t n o d e s a b s i d r i h t e h t ;t s e r e t n i s' r e s u h c a e f o s d e e n e h t o t d r a g e r t u o h ti w e m o s d n e m m o c e r , s r e s u f

o high populartiyusers ,these popular friends is through " s n a f " f o r e b m u n e h

t ,transmission raito, the leve lof acitvtiy to calculate ,Users t c e p x e t o n o d s d n e i r f d e d n e m m o c e r f o r e b m u n e g r a l a e v i e c e

r .Therefore ,the

o r c i m g n it s i x

e -blog recommendedapproachhascertain ilmtiaitons ,howto digou t r e s u e h t f o s t n e t n o c e h

t interes tfromthemassivemicro-blogdataandpersonailzed b s a h n o it a d n e m m o c e

r ecomeaho tspo tformicro-blogrecommendaiton. u n o d e s u c o f s d o h t e m d e d n e m m o c e r g n i g g o l b o r c i m l a n o it n e v n o

C ser interest ,

, s i s y l a n a s e r u t a e f d n a t n e t n o c g n i g g o l b o r c i

m ignoret hepotenita lilnkbetweensocia l o r c i m d e z il a n o s r e p f o e l o r e h t n i s r e s u k r o w t e

n -blog recommendaiton .In order to n o it c e n n o c f o d n i k s i h t y l p p

a reasonablyt othepersonailzedrecommendaiton ,Int he t s u r t f o e e r g e d e h t g n i r e d i s n o c e W , r e p a

p a ndtheuseri nteracitontrust ,ani mproved t s e r e t n i r e s u d e c u d o r t n i h c i h w , m h ti r o g l a n o it a d n e m m o c e r g n i r e tl i f e v it a r o b a ll o c e t a i v e ll a , t s u r t n o it c a r e t n i r e s u d n a y ti r a li m i

s data sparseness problem ,integrates . s n o it a d n e m m o c e r d e z il a n o s r e p n o it a m r o f n i n o it c a r e t n i d n a t n e t n o c g n i g g o l b o r c i m a d e z i n a g r o s i r e p a p e h t f o r e d n i a m e r e h

T s follows .Seciton 2 discusses g n i g g o l b o r c i m e h t s t n e s e r p 3 n o it c e S , k r o w d e t a l e r d e d n e m m o c e r g n i g g o l b o r c i m s l l a r e v o d e d n e m m o c e

r tructureandthedegree ofuserinteres tandtrus tcalculaiton f f e e h t d e i f i r e v t e s a t a d l a e r y b d n a s t e s a t a d e h t s e b i r c s e d 4 n o it c e S , p e t

s ec tof

d n e m m o c e r g n i g g o l b o r c i

m ,Finally ,wedrawt heconclusioni nSeciton5.

S K R O W D E T A L E R a w ] 1 [ m e t s y s d e d n e m m o c e

R s originally used in the field of e-commerce , n o it a d n e m m o c e r d e z il a n o s r e p , . c t e , s t s e r e t n i , s d e e n n o it a m r o f n i ' s r e s u o t g n i d r o c c a . r e s u e h t o t n o it a m r o f n i r e h t o d n a s t c u d o r p , n o it a m r o f n i d e t s e r e t n i l li w m e t s y s d e d n e m m o c e r e h t s i e r o c m e t s y s n o it a d n e m m o c e

R algortihm , the main

f o n o it a d n e m m o c e r e h t g n i r e tl i f e v it a r o b a ll o c : e d u l c n i s m h ti r o g l a n o it a d n e m m o c e r , s e l u r n o it a i c o s s a e h t n o d e s a b f o n o it a d n e m m o c e r e h t , t n e t n o c e h t n o d e s a b e v it a r o b a ll o c , m e h t g n o m A . c t e , s e i g o l o n h c e t g n i r e t s u l c d e s a b n o it a d n e m m o c e r i

f tlering algortihm [2] depend on tiems' score matrix of the targe tuser or simliar s i h t t u B . e u q i n h c e t r a l u p o p t s o m e h t s i , d n e m m o c e r o t s r e s u t e g r a t e h t f o s r e s u s s a l

c ic recommendaiton algortihm canno t be direclty used for user friends d e d n e m m o c e

r in socia lnetwork .In addiiton ,due to the data sparstiy of the socia l . l a e d i t o n s i g n i r e tl i f e v it a r o b a ll o c g n it s i x e f o s s e n e v it c e f f e e h t , k r o w t e n n i s m e t s y s d e d n e m m o c e r g n i g g o l b o r c i m , g n i g g o l b o r c i m f o e s u y l e d i w e h t h ti W t a e r g d e s u o r a s a h h c r a e s e r c i m e d a c a e h

t concern . Proposed learning to rank [

l e d o

m 3] ,selec tthe length of tweets ,qualtiy of the tweets ,influence of user three s t u p t u o h c i h w s t e e w t e h t k n a r o t d o h t e m a e s o p o r p r e p a p s i h T . r e tt i w T f o s e r u t a e f r i e h t n o d e s a b s t e e w t d e h c t a m e h

(3)

e h t r e d i s n o

c individua lneedsofusers .Chene tal[4] ,proposedamethodofmaking e d o m g n i k n a R e v it a r o b a ll o C n o d e s a b s n o it a d n e m m o c e r d e z il a n o s r e p t e e w

t l .

, s t e e w t f o s r o t c a f l e v e l c i p o t , s r e s u n e e w t e b s n o it a l e r d n a t n e t n o c r e tt i w t d e t a r g e t n I s r e s u e d i v o r p o t , k r o w t e n l a i c o s e h t n i e c n e u l f n i r e s u d n a s e r u t a e f t n e t n o c r e tt i w t r e s u d e z il it u y ll u f t o n d o h t e m e h T . tl u s e r s n o it a d n e m m o c e r d e z il a n o s r e p h ti w r e t n

i aciton informaiton , i t is recommended no t ideal . collaboraitve fitlering . d e c u d o r t n i s i y ti r a li m i s r e s u f o t p e c n o c e h t n o d e s a b m h ti r o g l a n o it a d n e m m o c e

r [5 ,]

y ti r a li m i s e v it c a r e t n i r e s u d n a y ti r a li m i s e t u b i r tt a r e s u e h t e t a l u c l a

C and ftiitng the

o s d n i k o w

t fsimliartiytogivetheulitmatesimliartiyoftheuser, recommendtarge t o r c i m n o d e s a b y l n i a m s i m h ti r o g l a e h t , s r e s u s t s e r e t n i r a li m i s h ti w s r e s

u -blog

e h T . h g u o n e e s i c e r p t o n s i y c a r u c c a n o it a d n e m m o c e r d n a y ti r a li m i s t n e t n o c r e s u d e s o p o r

p -based clustering algortihm[6] heterogeneous socia l network t n e t n o c , n o it c a r t x e e r u t a e f t n e t n o c g n i g g o l b o r c i m , n o it a d n e m m o c e

r -basedsimliartiy

s d n e i r f t n a v e l e r r o f s tl u s e r g n i r e t s u l c l a n i f e h t , p u o r g r e s u e r it n e e h t o t g n i r e t s u l c . d e d n e m m o c e

r To some extent ,the algorithm reduces the impac tof the matrix n o it a d n e m m o c e r f o y ti l a u q e h t n o s m e l b o r p t r a t s d l o c d n a y ti s r a p

s ,bu tthe users'

d e d n e m m o c e r r e w o l a t e s , g n i g n a h c y lt n a t s n o c e r a s t s e r e t n i d n a s e c n e r e f e r p , e g a r e v o

c Suchalgortihmshaveno tmadeaneffecitvewayt osolvet hisproblem. o r c i m g n it s i x

E -blog recommendaiton algortihm has achieved good resutls ,Bu t o r c i m f o e v it c e p s r e p e h t m o r f t r a t s y l n o m e h t f o t s o

m -blogconten toruserinterest , e

li h

w ignore the importanceof trus trelaitonship among therecommended objects . [7] proposed a nove lsocia lrecommended method and incorporate tiem relaitons , t n e t x e e m o s o t , r e s u t e g r a t e h t s a d e d n e m m o c e r x i r t a m p i h s n o it a l e r g n i s u g n it r a t s e s u l l u f e k a m t o n d i d t u b , m e l b o r p l a e d i t o n s i y c a r u c c a d e d n e m m o c e r e h t e t a i v e ll a c i m f

o roblogging users a variety of informaiton ,sparse matrix falied to solve the n e e w t e b p i h s n o it a l e r t s u r t d e d n e m m o c e r e h t m o r f e l c it r a s i h t , e r o f e r e h T . m e l b o r p o r c i m f o e c n e u l f n i e h t y d u t s o t , w e i v f o t n i o p t s e r e t n i r e s u d n a s t c e j b

o -blog

. s tl u s e r d e d n e m m o c e r R E T L I F E V I T A R O B A L L O

C INGALGORITHMBASEDONI NTEREST T S U R T D N A f o r e b m u n e h t , m e t s y s n o it a d n e m m o c e r d e z il a n o s r e p g n i g g o l b o r c i M s i s r e s u f o r e b m u n e h T . d e t a d p u y lt n e u q e r f o s l a t u b , e v i s s a m s i g n i g g o l b o r c i m r e s u , e l b a t s y l e v it a l e

r -based collaboraitvefitleringrecommendaiton ,relaitvelyeasy [ , r e p a p s i h t n I . s r e s u n e e w t e b y ti r a li m i s e t a l u c l a c o

t 8-9] analyze the users'

e t a d i d n a c t n a v e l e r t s o m e h t d n e m m o c e r s r e s u r o f n o it a t n e i r o s t s e r e t n i d e z il a n o s r e p g n it a r r e s u t i c il p x e o n s i e r e h t e s u a c e B . r e w o ll o

f data eb tweenmicro-blogplatform , o r c i m f o e s u e h

t -blogconten tand user interaciton informaiton ,the users interested r e s u t c u r t s n o c o t x i r t a m f o a t a d n o it a z il a it i n i s a e e r g e

d - topic feature matrix ,the

t a m a t a d n o it a z il a it i n i s a s n o it c a r e t n i r e s u a t a d t i c il p m

i rix ,buliduser- socia lfeature o r c i m n o d e s a b s r e s u n e e w t e b n o it a l e r r o c e h T , x i r t a

m -blog conten tand socia l

l a n a s i p i h s n o it a l e

(4)

s r e s u t e g r a t e h t t e g n e h t d n a , s r e s u f o y ti r a li m i s e v i s n e h e r p m o

c of neighbor set .

p o T e h t o t g n i d r o c c

A - Nsoritngalgortihmsproducerecommendaitonsfort heuser.

o r c i

M -blogU rs Ie nterest Calculaiton

r e s u e h t f o n o it c e l f e r e k il , s t n e m m o c , g n i d r a w r o f , k r o w t e n l a i c o s a n

I interest .

a f o n g i s e d e h t , n o it a m r o f n i s i h t h ti

W reasonableuseri nteres tmodelt ogett het opic o

r c i m f o y r o g e t a

c -blogusers,t ofurtheri nfert heusers'i nterestst endencies. 1

[ l e d o m c i p o

T 0] LDA as a kind of unsupervised tex t topic generaiton o

t d e s u e b n a c , d e i d u t s y l e d i w n e e b s a h l e d o m y ti li b a b o r

p idenitfylarge-scalese tof

1 [ n o it a m r o f n i c i p o t g n i y l r e d n u n i s u p r o c r o n o it c e ll o c t n e m u c o

d 1-12] .Int hispaper ,

e h t r e f n i d n a , n o it u b i r t s i d l a it n e t o p r e s u g n i n i m o t l e d o m c i p o t A D L d e v o r p m i e w

o r c i m f o n o it u b i r t s i d e h t h g u o r h t n o it a t n e i r o t s e r e t n i s' r e s

u -blogs'topic .In order to

e h t d n i f y l e t a r u c c

a interes ttopic of users ,this paper constructs microblogging o

r c i m d e t s e r e t n i l l a o t g n i d r o c c a t n e m u c o

d -blogs ,a user corresponding to the I

m e t s y s l a c i x e l d r o w e s e n i h C e h t e s u n e h t d n a , t n e m u c o d a f o s t n e t n o

c CTCLAS

e h t t e g o t ) m e t s y S s i s y l a n A l a c i x e L e s e n i h C , y g o l o n h c e T g n it u p m o C f o e t u ti t s n I (

e r o m s i r e s u e h t t a h t s n a e m t i , r e h g i h s i d r o w y e k f o y c n e u q e r f e h t , s d r o w y e k

e s n e s n o n , s d r o w p o t s , s n o c it o m e g n i g g o l b o r c i m e v o m e R . m r e t e h t n i d e t s e r e t n i

A D L e s u , s d r o

w topic mode lmining topic of each document ,ge tthe user-topic t e s s c i p o t , c i p o t f o n o it c e ll o c T t e s e r p a s t n e s e r p e r C t e L . x i r t a m

=

C � �C ,C� � …� C� � ,ti samicro-blog ,P� � �C |t indicatest hatt hemicro-blog t c

i p o t i C f o y ti li b a b o r p r o i r e t s o p e h t o t g n o l e

b ,The greater the value of P� �C |t) , e

h t t a h t s e t a c i d n

i possiblitiyofmicro-blog tbelongt ot het hemeofCii shigher. t

d e t s o p r e s u a e s o p p u

S he number of microblogging is d , make

� � � � �t t …� t� � represents microblogging colleciton of certain users ,the topic T

a e l b a li a v a s' r e s u e h t f o r o t c e v e r u t a e

f -dimensiona l vector

=

Z � � � � �v v …� vT� isdescribed ,asfollows:

v� = ( 1)

o r c i m y n a d e t s o p t o n s a h r e s u e h t n e h

W -blog ,the feature vectoriszero vector . .i

C c i p o t e h t n i d e t s e r e t n i s i r e s u e h t e r o m e h t ,i v t n e n o p m o c e h t r e h g i h e h T

s e r e t n I . 1 n o it i n if e

D tdegre e between th euse .r Le tthe se tof the user m

i S e n i s o c , } n u , … , 2 u , 1 u { =

U Ibetween two vectors defines as the degree of w

t e b y ti r a li m i s t s e r e t n

i eentheuseru iandthe userum. Theformulas arepresented .

s w o ll o f s a

m i

S I= = ( 2)

s t n e s e r p e r m Z , i Z , s r e s u o w t n e e w t e b y ti r a li m i s t s e r e t n i d e s s e r p x e I m i S e r e h W

(5)

c i

M or -blogU rs Te rus t Calculaiton

o l b o r c i

M ggingi nteracitoni nformaiton ,includingcomments ,forwarded ,ilkeand l a i c o s e h t f o h t g n e r t s e h t s t c e l f e r y lt c e r i d r o i v a h e b n o it c a r e t n I . s n o it c a r e h t o

e r o m e h t , s r e s u n e e w t e b t s u r t f o e e r g e d e h t , s i t a h t , s r e s u n e e w t e b s p i h s n o it a l e r

e r

f quenti nteracitonshowst hatt hemorei nitmaterelaitonship .

e e r g e d t s u r t e h t e z y l a n a o t s e r u t a e f l a i c o s f o s d n i k r u o f r e d i s n o c e w , r e p a p s i h t n I

. s p i h s n o it a l e r l a i c o s n o d e s a b

) 1

( Thedegreeofconcernoft heuser .Theformulasarepresentedasfollows.

Wf(uo)= ( 3)

f o r e b m u n e h t d n a s r e w o ll o f f o r e b m u n e h t t a h t o it a r e h t s e t o n e d ) o u ( f W e r e h W

, o u r e s u f o s e e w o ll o

f followers (uo) represents the number of fans of user uo , .

r e h s il b u p t s e r e t n i s r e s u f o r e b m u n e h t s t n e s e r p e r ) o u ( s e e w o ll o f

2( ) The number of users uo interes tin the followers of the user ut .The .

s w o ll o f s a d e t n e s e r p e r a s a l u m r o f

Wo(uo)=followees(uo) followers(ut |) ( 4)

b e h t , e r e H . m s i n a h c e m d a e r p s t s e r e t n i e h t s i a e d i c i s a b e h t ) 4 ( n o it a u q

E asic

s i e l p o e p g n o m a r o i v a h e b n r e c n o c e h t t a h t n o it p m u s s a e h t s i m s i n a h c e m d a e r p s

e h t w o ll o f a r e s u e h t f I . d a e r p s n a c e r u t a n n i t s e r e t n i s i h t d n a , t s e r e t n i e h t n o d e s a b

b r e s u e h t n e h t , c r e s u w o ll o f b r e s u e h t f i ; b r e s u n i d e t s e r e t n i a r e s u t a h t b r e s

u plays

e d u l c n o c o t e l b a d n a , c r e s u d n a a r e s u n e e w t e b r o t a c i n u m m o c t s e r e t n i f o e l o r e h t

e h t s a h c u s t s e r e t n i f o r e b m u n e h t e r o m e h t , s r e s u o w t r o F . c r e s u n i d e t s e r e t n i a r e s u

. c r e s u r o f a r e s u e h t d e t s e r e t n i e r o m e h t n e h t , s r o t a c i n u m m o c b r e s u

u n e h T ) 3

( mberofcommonfolloweeofuseru tanduseruo ,calculatedas(5)

Wt(uo)=followees(ut) followees(uo) | ( 5)

s e e w o ll o f , r e h s il b u p n i t s e r e t n i o u f o r e b m u n e h t s e t o n e d ) o u ( s e e w o ll o f e r e h W

e r e t n i t u r e b m u n e h t s e t o n e d ) t u

( s tinpubilsher .Obviously ,thenumberofcommon .

p i h s n o it a l e r e t a m it n i e r o m e h t , o u r e s u e h t d n a o u r e s u e h t f o s d n e i r f

e h t r e h g i h e h t , n r e c n o c f o e e r g e d e h t r e h g i h e h t , r e s u e h t f o e c n e u l f n i e h T ) 4 (

, d e d r a w r o f s a w g n i g g o l b o r c i

m comments ,ilke .Asformula )( 6

u ( c

W o)=CountR(uo)+CountC(uo)+CountL(uo) ( 6)

h t s e t o n e d ) o u ( R t n u o C e r e h

W e number ofcommentsof user uo ,CountC(uo) ) o u ( L t n u o C , g n i g g o l b o r c i m o u r e s u d e d r a w r o f y b f o r e b m u n e h t t a h t s e t a c i d n i

. e k il y b o u r e s u e h t f o r e b m u n e h t s e t o n e d

o it i n if e

D n 2. Interacitve trus tbetween users Le tthe user se tU = {u1 ,u2 ,... , t

u r e s u n o d e s a b t s u r t e v it c a r e t n i o u r e s U } n

(6)

m i

S R(uo)=wfWf(uo) w+ oWo(uo) w+ tWt(uo) 1+ -w( f-wo-wt)Wc(uo) ( 7)

n r e c n o c f o e e r g e d e h t s e t a c i d n i ) o u ( f W e r e h

W ,Wo(uo)representsthenumber o u r e s u t u r e s u e h t t a h t s n a e m ) o u ( t W , t s e r e t n i f o t c e j b o o u r e s u n a f t u s r e s u f o

f o e c n e u l f n i e h t s t n e s e r p e r ) o u ( c W , s d n e i r f f o r e b m u n n r e c n o c n o m m o

c featuresof

. r e s u e h t

e h t d n a a r e s u n e e w t e b s d n e i r f w o ll o f n o m m o c n e t e r a e r e h t f

I userb ,bu tonlya

, c r e s u d n a a r e s u n e e w t e b d n e i r f w o ll o f n o m m o

c obviously ,user a is more ilkely b

r e s

u than user c. Visible ,The number of common interests between users can .

s r e s u n e e w t e b n o it a l e r r o c t s e r e t n i y f it n a u q

o w , f w s t h g i e w e h t ) 7 ( n o it a u q

E and wr direclty affec tthe recommendaiton m

it s e y l e t a r u c c a e r o m o t r e d r o n I . n o it a l u c l a c e r o c

s atethevaluesofwf ,woandwr , e l b a i r a v t n e d n e p e d e h t f o n o i s s e r g e r r a e n il , l e d o m n o i s s e r g e r r a e n il a s a ) 7 ( a l u m r o f

a s e l b a i r a v t n e d n e p e d n i r u o f , ) o u ( R m i S =

y rex1=Wf(uo) ,x2=Wo(uo) ,x3=w t e

r e w s t n e i c i f f e o c g n i d n o p s e r r o c t n e m u g r a e h t , ) o u ( c W = 4 x d n a ) o u

( = wf , =

, o

w = w tand = - f1w - ow -wt .Suppose there are n sets of observaitons can be d

e t n e s e r p e

r by the formula (7) to derive an expression for the ilnear regression :

) 8 ( n o it a u q e

( 8)

s s o l g n i c u d o r t n i d n a l e d o m n o i s s e r g e r r a e n il e h t o t n i s n o it a v r e s b o f o t e s h c a E

o c e u l a v m u m i n i m g n it t e g n o it c n u f s s o l e h t g n i v l o s n e h w , ) 9 ( n o it c n u

f rresponding

e h t o

t � � dan 4 .

y ( = ) ) 4 x , 3 x , 2 x , 1 x ( f , y (

L -f(x1,x2,x3,x4))2 ( 9)

d e t a r g e t n

I Interes tandTrus tRecommendaitonAlgortihm

e f f a s n o it a d n e m m o c e r r e s u , s e c i v r e s g n i k r o w t e n l a i c o s g n i g g o l b o r c i m r o

F c t

r e s u d n a n o it a m r o f n i t n e t n o c s' r e s u e h t : n o it a m r o f n i f o s e p y t o w t y l n i a m e r a s tl u s e r

f o y ti r a li m i s e h t n e v i g e r a s r o t c a f o w t e s e h t r e d i s n o C . n o it a m r o f n i n o it c a r e t n i

: s w o ll o f s a ) 0 1 ( n o it a u q e n i l e d o m n o it a l u c l a c r e s u e h t n e e w t e b s t s e r e t n i

= m i

S (10)

m i S e h t e r e h

W indicates users comprehensive simliartiy ,integrated simliartiy t

n e t n o c : s t r a p o w t f o s t s i s n o

c -based impac tfactorSim Iandtheimpac tfactor SimR s

a b ) o u

( edonsocia lrelaitonships,t heweigh toft hei mpac tfactori s .Togeneratea d

e d n e m m o c e r f o t e s d o o

g , the algortihms should be combined wtih the rea l .

(7)

m h ti r o g l a e h t , n o i s s u c s i d e v o b a e h t m o r

F I&R_CF (Interes tAnd Relaitonship e

v it a r o b a ll o

C -Fitlering)gives the procedureofgeneraitngtherecommendedse tfor .

o u r e s u e h t

) N t n i , U , o u ( F C _ R & I 1 m h ti r o g l A

. N r e b m u n d e d n e m m o c e r e t a r e n e g , U t e s r e s u e t a n r e tl a , o u r e s U : t u p n I

o c e r o u s r e s U : t u p t u

O mmendaitonse tC. n

i g e B

� 1� Simliartiycalculaiton 1

( -1)Theuseri nteres tcalculaiton.

LDAtopic mode lminingtopicofeachdocumen taccordingtothemicro-blog r

e s u e h t t e g , t e

s -topicmatrix .Wecalculatetheuserfeaturevectoraccordingtouser e

r e t n

i s tsimliartiySim.I 1

( -2)Userst rus tcalculaiton.

① esitmated tha tweights wf ,wo ,and wr ,randomly selected n users from the t

e s a t a

d ,use t hehistoricali nteracitondataasat rainingse tforeachuseraccordingt o .

) 7 ( a l u m r o f

② usingequaiton(3) ,(4) ,(5) ,(6)tocalculatetheweightsofWf(uo) ,Wo(uo) , .

) o u ( c W d n a ) o u ( t W

③ usingthe ilnearregressionmodelt hrough tieraitveweightedvalueofwf ,wo .

r w d n a

④ basedonuserinteracitoninformaiton ,calculatefoursocia lfeaturesWf(uo) , o

u ( o

W ) ,Wt(uo)andWc(uo) inthemicrobloggingsocia lnetwork .The four kinds t s u r t s' r e s u e h t e t a l u c l a c o t g n it ti f r a e n il y b d e t a l u c l a c e r e w s e u l a v e r u t a e f l a i c o s f o

, g n i k n a r y ti r a li m i s e v i s n e h e r p m o c s' r e s u e h t o t g n i d r o c c

A g et the fina l

, L t e s n o it a d n e m m o c e

r In accordance wtih the Top-N sor talgortihm to generate .

s r e s u r o f s n o it a d n e m m o c e r

. C t e s n o it a d n e m m o c e r o u s r e s U : t u p t u O

d n E

o t m h ti r o g l a n o it a d n e m m o c e r e e r g e d t s u r t d n a t s e r e t n i r e s u e h t n o d e s a B

tl u s e r e h T . y ti r a li m i s e v i s n e h e r p m o c s' r e s u e h t e t a l u c l a

c softhetwo comprehensive

r e s u N p o t e h t e k a t n e h T . s tl u s e r n o it a d n e m m o c e r l a n i f e h t s a g n i k n a r

. s r e s u t e g r a t e h t o t s n o it a d n e m m o c e r

I S Y L A N A L A T N E M I R E P X

E S

l a t n e m i r e p x

E Environmen tandT eh Dataset

c e l e s , d o h t e m d e s o p o r p e h t f o s s e n e v it c e f f e e h t y f i r e v o

T ttherea lexperimenta l

l a it n e u l f n i t s o m e h t s i g n i g g o l b o r c i m a n i S . m r o f t a l p g n i k r o w t e n l a i c o s m o r f t e s a t a d

e h t , r e p a p s i h t n I . n o o s d n a s d n e i r f w o ll o f , s t n e m m o c n a c s r e s u , m r o f t a l p e c i v r e s

c i s a b s' r e s u t e g e r a w t f o s r e l w a r c g n i g g o l b o r c i m e s u l a t n e m i r e p x

e informaiton ,users

o r c i

(8)

h t d a e r b e h t g n i s

u -firs talgortihm ,collecitng20,000userdata .Inordert oi mprovet he e

t n i e h t , g n i n a e l c y b , m h ti r o g l a t s e t e h t f o y c a r u c c

a graitonofdatasets ,obtained5424 o

r c i m 7 2 6 6 7 2 g n i d u l c n i , s r e s

u -blogs, 115583i nteracitoni nformaiton .Int hispaper , s

s o r

c -overstudy ,thedataseti sdividedi nto80%t rainingsetsand20%tes tsetsand .

m h ti r o g l a e h t e t a d il a v o t s t n e m i r e p x e f o s t e s e l p it l u m y b

o v o n e L e h t n o g n i n n u r e d o c t n e m i r e p x e e h T . a v a J d n a n o h t y P n i e d o C

. 8 . 1 n o i s r e v k d j , 1 . 5 . 3 n o h t y P n o i s r e v n o h t y P , 0 6 4 E d a P k n i h T

: m h ti r o g l a e c n e r e f e r l a t n e m i r e p x

E

p o T n o e s a

B -K simliar collaboraitve fitlering recommendaiton algortihm e

v it a r o b a ll o C

( -Fitlering ,CF)tes talgortihm basedon opensource reference ilbrary t

u o h a M e h c a p

A machine learning implementaiton .CF based on user ,the user it

a l e r r o c s' n o s r a e P g n i s u e t a l u c l a c y ti r a li m i

s oncoefficient ,thefina lresul toftheuse p

o T n o d e s a b d e d n e m m o c e r f

o -Krecommended.

n o it a u l a v

E Metrics

t e s u e w , r e p a p s i h t n

I hreepopular evaluaitonmetrics :Precision ,Recal land F .

e r u s a e m

t a d n e m m o c e r e v it c e f f e n i s tl u s e r d e d n e m m o c e r a s i n o i s i c e r

P ion accountedfor

e h t f o s s e n e v it c e f f e e h t t c e l f e r n a c , d e d n e m m o c e r l l a f o n o it r o p o r p e h t

: s w o ll o f s a ) 1 1 ( a l u m r o f n o it a l u c l a c e h t , m h ti r o g l a n o it a d n e m m o c e r

n o i s i c e r

P = (11)

e r e h

W Nrs is the number of correc trecommended in recommended ilst ,Nr is .

s tl u s e r d e d n e m m o c e r f o r e b m u n e h t

n o it r o p o r p e h t r o f d e t n u o c c a s d n e i r f n i d e t s e r e t n i s r e s u d e d n e m m o c e r a s i ll a c e R

: s w o ll o f s a ) 2 1 ( a l u m r o f n o it a l u c l a c e h t , s d n e i r f s i h f o l l a f o

= ll a c e

R (12)

N e r e h

W r s is the number of correc trecommended in recommended ilst ,Nr is .

r e s u e h t f o s d n e i r f l a t o t f o r e b m u n e h t

h c i h w , e t a r l l a c e r d n a n o i s i c e r p f o n a e m c i n o m r a h d e t h g i e w a s i e r u s a e m F

r e tt e b e h t e r u s a e m F r e h g i h e h T . s e t a r l l a c e r d n a n o i s i c e r p f o s tl u s e r e h t s e n i b m o c

e r d e d n e m m o c e r e h

t sutls.

F measure= (13)

p o T f o r e b m u n d e d n e m m o c e r e h t , t n e m i r e p x e s i h t n

I -N ofN = 2,4,6,8,10 ,five

(9)

l a t n e m i r e p x

E Resutl sandAnalyssi

f o s e u l a v g n i h g i e w e h t e n i m r e t e d o T . 1 n

I order to calculate the weigh tα ,according to rea ltes tdatase tleas tsquares t s e b e h t n i a t b o o t α f o e u l a v e h t t s u j d a o t e u n it n o c , g n i m m a r g o r p r a e n il d n a g n it ti f

F 1 e r u g i F n i n w o h s s A . s tl u s e

r measure curve comparison ,when the maximum ,

4 . 0 s i α f o e u l a

v F measuret aket heopitma lpersonailzedrecommendaitonresul.t

g i

F u 1. re algortihm andFmeasure.

e r u g i

F 2 .algortihmprecisioncomparisonchart. Figure3 .algortihmcontras tfigurerecal lrate.

i s e v i s n e h e r p m o c h ti w e c n a d r o c c a n i n o it a d n e m m o c e r l a e r n

I mliartiy

f o e u l a v e h t , e r o f e r e h T . tl u s e r d e d n e m m o c e r s a N t s r i f e h t t c e l e s , r e d r o g n i d n e c s e d

. ll a c e r d n a n o i s i c e r p d n e m m o c e r s d n e i r f g n i g g o l b o r c i m p i h s n o it a l e r t c e r i d a s a h N

n e v e d n a , g n i g g o l b o r c i m f o r e b m u n t n e r e f f i d a e v i e c e r n a c r e s u h c a e e s u a c e b d n A

wli lbeverydifferent ,tii sno tpossiblesimplyt oNt oagivenvalue. N

n e h

w ≤4 ,wtih the increase o f the length of the recommendaiton ilst ,the s

i n o i s i c e r

p increasing, thispaperalgortihmwtihhighe raccuracythanthebasedon n

e m m o c e r g n i r e tl i f l a i c o

(10)

e s u a c e b s i s i h T . e n il c e d o t e u n it n o c e t a r y c a r u c c a e h t , s e s a e r c n i t s il d e d n e m m o c e r

n e c s e d y ti r a li m i s e v i s n e h e r p m o c e h

t ding order, there wli lbe some lesser interes t .

e t a r y c a r u c c a d e c u d e r o t d a e l ,t s il n o it a d n e m m o c e r e h t o t n i g n i g g o l b o r c i m

e h t n a h t ll a c e r r e h g i h s i d o h t e m d e s o p o r p e h t , e t a r l l a c e r e h t t a h t s w o h s 3 e r u g i F

r e p a p s i h t , 6 = N n e h w , m h ti r o g l a g n i r e tl i f l a i c o s n o d e s a

b algortihmto achievethe

s d o h t e m h t o b h g u o h tl a , 3 d n a 2 s e r u g i F m o r f e e s n a c e w s A . % 3 . 5 6 f o l l a c e r t s e h g i h

f o r e b m u n e h t e s a e r c n i o t n o it a d n e m m o c e r d o o g a h ti w e t a r l l a c e r d n a n o i s i c e r p e h t

n o i s i c e r p e h t n i d o h t e m d e s o p o r p e h t t u b , e m a s e h t y l h g u o r s d n e r

t and recal lrate

r e tt e b a s a h d o h t e m s i h t s w o h s h c i h w , m h ti r o g l a F C e h t n a h t r e tt e b y lt n a c i f i n g i s s a w

. e c n a m r o f r e p

d e s o p o r p e h t t a h t d e d n e m m o c e r g n i g g o l b o r c i m t a h t w o h s s tl u s e r , ll a r e v O

p m i , t n e t x e n i a t r e c a o t , t c e f f e n o it a d n e m m o c e r r e tt e b a s a h m h ti r o g l

a rove the

. s m h ti r o g l a s d n e i r f d e d n e m m o c e r g n i g g o l b o r c i m e h t f o y ti l a u q

U T U F D N A S N O I S U L C N O

C REWORK

d n i f y li s a e o t s r e s u p l e h n a c m e t s y s d e d n e m m o c e R g n i g g o l b o r c i m d e z il a n o s r e P

o f n i e h

t rmaiton they are interested in . This paper considers the users' own r

e f e r

p ences and interests microblogging users interac twtih trust ,we pu tforward . g n i r e tl i f e v it a r o b a ll o c d e v o r p m i n o d e s a b d o h t e m d e d n e m m o c e R g n i g g o l b o r c i m

l a n o it i d a r t e h t o t y ti r a li m i s t s u r t d e d n e m m o c e r d n a t s e r e t n i s' r e s u e h t h ti w d e n i b m o C

p o T e h t h ti w e c n a d r o c c a n i ; d e v o r p m i s a w m h ti r o g l a g n i r e tl i f e v it a r o b a ll o

c - N

l A g n it r o

S gortihm for the user to generate a recommendaiton ilst .Selec tthe user g n i g g o l b o r c i m g n il b a n e , r e s u e h t o t d e d n e m m o c e r d n e i r f a t s e r e t n i o t y l e k il t s o m

d e z il a n o s r e p y f i r e v o t a t a d t s e t l a e r e h t , y ll a n i F . s n o it a d n e m m o c e r d e z il a n o s r e p

c a r u c c a n o it a d n e m m o c e

r y and efficiency of the method proposed in this paper . l

a e r g n i g g o l b o r c i m r e d i s n o c d l u o c k r o w e r u t u

F -itmepersonailzedrecommendaitons , .

y c n e i c i f f e e v o r p m i r e h t r u f d e d n e m m o c e r

S T N E M G D E L W O N K C A

y b d e t r o p p u s s i k r o w s i h

T the periodica lachievementsof the subjec tof pubilc (

y ti s r e v i n U u o h z g n e h Z n i a i d e m w e n f o n o it a c i n u m m o

c projec tXMTGGCBJSZ05) ,

( e c n i v o r P n a n e H f o t c e j o r p y g o l o n h c e t d n a e c n e i c

S projec t 144300510007) ,

s t c e j o r p h c r a e s e r y g o l o n h c e t d n a e c n e i c s y ti c u o h z g n e h

Z (projec t141PPTGG368).

S E C N E R E F E

R

.

1 We i Chen , Hsu W., Lee M. .L Modeilng 2013. User’s Recepitveness Over Time for .

e c n e r e f n o C R I G I S M C A h t 6 3 e h t f o s g n i d e e c o r P n o it a d n e m m o c e

(11)

.

2 ChaochaoChen ,JingZeng ,XiaoilnZheng ,DerenChen .2013.RecommenderSystemBasedon it

a l e R t s u r T l a i c o

S onshipsProceedingsofIEEEt he10thInternaitona lConferenceone-Business 3

1 0 2 , E E E I . g n i r e e n i g n

E .

.

3 Duan Y. ,Jiang L. ,Qin T. ,e tal .2010.An empirica lstudy on learning to rank of tweets e

c n e r e f n o C l a n o it a n r e t n I d r 3 2 e h t f o s g n i d e e c o r

P onComputaitona lLinguisitcs .IEEE,2010. .

4 ChenK. ,ChenT. ,ZhengG. ,e tal .2012.Collaboraitvepersonailzedt wee trecommendaiton[C]/ / n i t n e m p o l e v e D d n a h c r a e s e R n o e c n e r e f n o C R I G I S M C A l a n o it a n r e t n I h t 5 3 e h t f o s g n i d e e c o r P

.l a v e i r t e R n o it a m r o f n

I ACM ,2012. .

5 R go n Huigui ,H ou Shengxu ,Hu Chunhua,MoJinxia .2014,Usersimliartiy-basedcollaboraitve li

f teringrecommendaitonalgortihmJ. Journa lonCommunicaitons ,35(2) :16- 42 .(inChinese) .

6 C nh e Kehan , Ha n Panpan , Wu Jian. 2011. User clustering based socia l network a

d n e m m o c e

r itonJ. ChineseJourna lofComputers ,36(2) :349- 935 .(inChinese) .

7 G ou Lei ,Ma Jun ,Ch enZhumin, JiangHaoran.2014. IncorporaitngItemRelaitonsforSocia l n

o it a d n e m m o c e

R [J].ChineseJourna lofComputers ,37(1) :219- 822 .(inChinese) .

8 RamageD. ,DumaisS. ,e ta.l2010Characterizingmicroblogswtihtopicmodels .Proceedingsof 0

1 0 2 , I A A A , a i d e M l a i c o S d n a s g o l b e W n o e c n e r e f n o C I A A A l a n o it a n r e t n I e h

t .

.

9 NaveedN ,GottronT. ,KunegisJ. ,e tal .2011.Badnewstrave lfast :acontent-basedanalysisof g

n it s e r e t n

i nessont witter .Proceedingsoft heACMWebSci’11.ACM ,2011. .

0

1 Xu Ge ,Wa ng Houfeng. 2011. The Developmen to fTopics Models in Natura lLanguage J

g n i s s e c o r

P .ChineseJourna lo fComputers ,34(8) :1423-1435 .(inChinese) .

1

1 Chen Wenbin� Yang Chikai� Huang Yuanhao. 2011. Energy Saving Cooperaitve Spectrum .

J m e t s y S o i d a R e v it i n g o C r o f r o s s e c o r P g n i s n e

S IEEETransacitonsonRegularPapers ,58(4) : 1

1 7 - 372 �

. 2

1 SinghA.� BhatnagarM.R.� MalilkR.K .e tal� 2012.CooperaitveSpectrumSensinginMulitple U

k r o w t e N o i d a R e v it i n g o C d e s a B a n n e t n

A sing an Improved Energy Detecto J� IEEE )

1 ( 6 1 , s r e tt e L s n o it a c i n u m m o

References

Related documents

The nanohardness and elastic moduli (Graph 1) of two bulk fill and two incremental fill resin composites evaluated in this study showed Filtek Bulk Fill with good hardness

Gebrehiwot et al BMC Pregnancy and Childbirth 2012, 12 113 http //www biomedcentral com/1471 2393/12/113 RESEARCH ARTICLE Open Access Making pragmatic choices women?s experiences

Although had no effect on the png phenotype, but crossing eight the eIF-5A protein was identified by its ability to promote extra copies of the wild-type cyclin B3 gene into the

STRUCTURE USING LIGHT WEIGHT (NATURAL PUMICE

At the same time, it is having its specialties like different authentication, privacy issues, issues to do with access control network configuration, information

SVR12 results from the Phase II, open-label IMPACT study of simeprevir (SMV) in combination with daclatasvir (DCV) and sofosbuvir (SOF) in treatment-naïve and -experienced patients

This study aimed a) to develop a questionnaire for as- sessing women’s attitude on medical ethics application in labor and delivery; and b) to assess the validity and reli- ability

: Assessment of pharmacological strategies for management of major depressive disorder and their costs after an inadequate response to first-line antidepressant treatment in