Database Reference
In-Depth Information
Rating Matrix T
U
User U
Item I
Social Network G
I
R
1
2
5
2
2
4
User Preference
Inference Engine
Item Likability
Inference Engine
Homophily
InferenceEngine
Pr(R I | a U )
Pr ( R UI | r VI :
Pr(R U | a ' I )
" V Œ U (I) « N (U) )
Aggregator
Immediate Friend
Inference Engine
Pr
(
R UI | a ' I , a U , r VI :
" V Œ U (I) « N (U)
)
Distant Friend
Inference Engine
" V ΠU (I)
Pr ( R UI | a ' I , a U , r VI
:
)
Updated Rating Matrix
T
U
I
R
1
2
5
1
3
4
2
1
3
Fig. 4.3 The architecture of a social network-based recommender system
influenced not only by his/her immediate friends; his/her distant friends can
also exert their influence indirectly through his/her immediate friends. Therefore,
SNRS incorporates these two types of influences, but it deals with them differently.
In particular, the immediate friend inference engine focuses on exploiting the
homophily effects among immediate friends, and the distant friend inference engine
leverages the immediate friend inference engine to bring homophily effects among
distant friends into consideration.
More specifically, the immediate friend inference engine contains four smaller
components: (1) User preference inference engine computes the probability distri-
bution of a target user U' s rating based on U' s preferences to the items similar to a
target item I ; (2) Item likability inference engine computes the probability distribu-
tion of the rating that item I receives based on the characteristics of the reviewers
similar to user U ; (3) Homophily inference engine utilizes homophily effects among
Search WWH ::




Custom Search