Document Sample
precision Powered By Docstoc
					Mao Ye , Peifeng Yin, Wang-Chien Lee, Dik-Lun Lee
       Pennsylvania State Univ. and HKUST

                    SIGIR 11
 Introduction and Motivation
 Model
 Experiments & Evaluation
 Conclusions
 Location Based Social Network(LBSNs): Foursquare,
  Gowalla, Brightkite, Loopt etc.
 Allow share tips or experience of Point-of-
  Interest(POIs) e.g. restaurants, stores, cinema
  through check-in behaviors
Main Elements in LBSNs
 Recommend new POIs to users can help them explore
  new places and know their cities better
 In LBSNs, different from other systems, “cyber”
  connections among users as well as “physical”
  interactions between users and locations captured in
  the systems, thus POIs recommendation in LBSNs is
  promising and interesting
 The idea of incorporating the geographical influence
  between POIs has not been investigated previously
 Three important factors:
    Geographical influence
    User preference of POIs
    Social Influence
 A fusion framework combine all three
Geographical Influence
 measures how likely two of a user’s
  check-in POIs within a given distance
 User power law distribution to model
  the check-in probability to the distance
  between two POIs visited by the same
  users: y  axb
 Given user i and his check-in history Li,
Geographical Influence
 Then for a new location lj , we have the probability for
  user I to check in lj as follows:
User-based CF
 Based on user similarity

    is the predicted check-in probability.
    is the similarity of user i and user k, and
 computed as follows:
Friend Based CF
 Based on recommendation from friends

   Friends have closer social tie
   Friends show more similar check-in bahavior
Fusion Framework
 Combine all of the three factors
Data Set
Performance Metrics
 Mark off some POIs and the systems return top-N
  recommended POIs
 Mainly examine below two metrics
   The ratio of recovered POIs to N, precision@N
   The ration of recovered POIs to the total POIs which are
    marked off , recall@N
 Model in this paper denotes as USG
   U for user preference
   S for social influence
   G for geographical influence

 Compared Methods
   User-based CF (U) : set α=β=0
   Friend-based CF (S): set α = 1, β=0
   GI-based (G): set α = 0, β=1
   Random Walk with Restart(RWR)
   User preference/social influence based (US): set β=0
   User preference/geographical influence based(UG): set α = 0
Tuning Parameters
 User preference plays a dominate role in contributing
  to the optimal recommendation
 Both social and geographical influence are innegligible
Performance Comparison Result
 USG always the best
 RWR may not be suitable for POI recommendation
 Social influence and geographical influence can be
 utilized to perform POI recommendation
Study on Item-based CF
 Regard POIs as “items” and denotes as L , and combine
  it with user preference(U) and geographical
 L brings no advantage at all in enhancing U or L in
  POI recommendation
   POIs in LBSNs not have been visited by sufficient users
Study on Social Influence
 User check-in behaviors and the user similarity
  calculated based on RWR
 Check-in behaviors and social tie strength
 The similarity in friends’ check-in behaviors not
  necessarily be reflected through social tie strength
Impact of Data Sparsity
 The larger the mark-off ratio x is , the sparser the user-
  Check-in matrix is
 Geographical plays an extremely important role when
  data is very sparse.
Test for Cold Start Users
 Consider users who have less than 5 check-ins after
  mark off 30%
 For cold start users, user preference is hard to capture,
  thus U performs bad , and as few check-ins, G also
  affects, and S is more useful in this situation
 First incorporate geographical influence into POI
 Incorporate U,S,G into a fusion framework
 Experiments conclusions
   Geographical influence shows a more significant impact
    than social influence
   RWR may be not suitable for POI recommendation,
    friends’ taste is different( friends have low common
    check-in ratio)
   Item-based CF is not effective
Future Work
 Combine semantic tags , e.g. location categories such
  as Store, Restaurants
 Combine geographical influence into Matrix
  Factorization Method
 Take location transition sequence into consideration
Thanks 

Shared By: