Fused Matrix Factorization with Geographical and Social Influence in Location-based Social Networks Motivation Ø Geographical influence is important in POI recommendation Ø Model user check-in behavior to help improve recommendation
Fused Matrix Factorization with Geographical and Social Influence in Location-based Social Networks Motivation Ø Geographical influence is important in POI recommendation Ø Model user check-in behavior to help improve recommendation Multi-center Gaussian Model: The prob. of visiting a location l is defined as: Centers are found by using greedy algorithm. Fused Framework: PMF: Model user preference MGM: Model geo-influence FMFMGM: Fused PMF and MGM normalized effect of check-in frequency
Where You Like to Go: Next Successive Point-of-Interest Recommendation Motivation Ø Time-aware POI recommendation is an important task Ø Two main properties in LBSNs: Ø Personalized Markov chain Ø Localized region constraint
Example 0. 85 1. 0 0. 70 0. 15 0. 20 0. 30 1. 0 0. 79 0. 59 0. 01 User 1 0. 40 Localized Region Constraint User 2
Model • FPMC-LR is to recommend a successive personalized POI by the prob. a user u will visit at time t: • Base on first-order Markov chain property Prob. for user u from location i to l
Model • FPMC-LR only consider the neighborhood locations of previous check-ins • Thus our FPMC-LR yields a transition tensor – Note: is reduced largely compared to around 100 when d = 40 km ,
Model • Model top-k recommendations as a ranking over locations: • The MAP estimator is • Learning algorithm: Stochastic gradient descent