Beruflich Dokumente
Kultur Dokumente
[1] Bao, J., Zheng, Y., & Mokbel, M. F. (2012, November). Location-based and preference-aware recommendation using sparse geo-social networking data.
InProceedings of the 20th International Conference on Advances in Geographic Information Systems (pp. 199-208). ACM.
[2] http://research.microsoft.com/en-us/projects/lbsn/
Introduction
• LBSN rating matrix => {user-check-ins} tuples
• Is rating/check-in frequency alone enough?
• Rating matrix often sparse => cold start
problem
• Need to focus beyond just the rating/check-in
frequency
• Can we utilize LBSN features?
Introduction..
• LBSN has unique features
– Different aspects (time, location category, social
relation, location distance etc) associated with
LBSN
• Challenges for efficient prediction
– Social information not always reliable
– 96% of people share < 10% of the commonly
visited places and 87% of people share nothing at
all [1]
– Wise fusion of aspects is needed
• Integration of multiple aspects for efficient
recommendation can be effective
[1]
M. Ye, P. Yin, and W.-C. Lee, “Location recommendation for location-based social networks,” in Proceedings of the 18th
SIGSPATIAL International Conference on Advances in Geographic Information Systems. ACM, 2010, pp. 458–461.
Motivation
Motivation..
• Geographical
– Preference to closest location
• Temporal
– Temporal popularity of POI
– Temporal check-in trend of user
• Categorical
– Locations with same category can be an option
• Social
– Influence of Friends/Followers-Followee
Related Research
• Yuan et al. [7] designed a model with the spatial
and the temporal aspect.
• Used the cosine similarity of check-in profiles.
• Defined the recommendation score for a user-
location tuple in terms of the aggregate of the
visits count on that location over all the users in
the dataset.
• Time constrained => same location and at the
same check-in time.
• No social and categorical aspect
Related Research..
• Ye et al. [1] fused the social and the spatial
aspects
• Used the willingness factor and the weighted
cosine similarity measure to compare the user
profiles for the recommendation
• The categorical and the temporal aspects were
not explored in their proposed model
Related Research..
• GeoMFTD [4] fused the spatial and the temporal
influence but still didn’t incorporate other major
aspects for the recommendation.
• For each POI i, used the average time spent by each
user to reach the POI j (j ε gl, where gl is the lth
geographical grid/region) from the POI i.
• Repeat for every user who had at least one check-in
at POI i and more recent one at the POI j into gl.
• Temporal aspect was incorporated by the temporal
coefficients to the POI influence.
Methodology
• Matrix Factorization characterizes both items
and users by vectors of factors inferred from
the user-item rating matrix
• user factor matrix U =[u1; u2...; um]
• item factor matrix V= [v1; v2....; vn]
• the approximation of the rating matrix R can
be defined as: R~UTV and regularized as:
Methodology..
• Map the user-location check-in frequency
matrix (R) into low dimension latent space:
• Regularized:
• Recommendation relation:
Dataset[1] statistics
Attributes Gowalla Weeplaces
[1] X.
Liu, Y. Liu, K. Aberer, and C. Miao, “Personalized point-of-interest recommendation by mining users’ preference
transition,” in Proceedings of the 22nd ACM international conference on Conference on information & knowledge
management. ACM, 2013, pp. 733–738.
Top 5 popular categories
Gowalla Weeplaces
Ye et al. [1] @5= 0.03030 @5= 0.00080 FMFMGM @5= 0.05900 @5= 0.00489
@10= 0.02300 @10= 0.00090 [5] @10= 0.06800 @10= 0.00687
@15= 0.01910 @15= 0.00100 @15= 0.08700 @15= 0.00873
LBSNRank @5= 0.08530 @5= 0.00060 GeoMFTD @5= 0.07719 @5= 0.00641
[2] @10= 0.08480 @10= 0.00060 [4] @10= 0.08947 @10= 0.00824
@15= 0.40900 @15= 0.00300 @15= 0.11578 @15= 0.00924
Wang et al. @5= 0.04490 @5= 0.00140 GeoTecs @5= 0.28400 @5= 0.00950
[3] @10= 0.04140 @10= 0.00207 @10= 0.36500 @10= 0.00920
@15= 0.04070 @15= 0.00220 @15= 0.38800 @15= 0.02770
Conclusion and Future direction
• Fused the (a) the geographical/spatial, (b) the
categorical, (c) the temporal and (d) the social
aspects into a POI recommendation model
• Future direction:
– Fusion of other aspects
– Context-based location recommendation
Acknowledgements
• The work was supported in part by the
National Science Foundation under grants
CNS-1126619, IIS-1213026, and CNS-1461926,
the U.S. Department of Homeland Securitys
VACCINE Center under Award Number 2009-
ST-061-CI0001, and a gift award from Huawei
Technologies Co.Ltd.
References
1. M. Ye, P. Yin, and W.-C. Lee, “Location recommendation for location-based social networks,” in
Proceedings of the 18th SIGSPATIAL International Conference on Advances in Geographic
Information Systems. ACM, 2010, pp. 458–461.
2. Z. Jin, D. Shi, Q. Wu, H. Yan, and H. Fan, “Lbsnrank: personalized pagerank on location-based social
networks,” in Proceedings of the 2012 ACM Conference on Ubiquitous Computing. ACM, 2012, pp.
980–987.
3. H. Wang, M. Terrovitis, and N. Mamoulis, “Location recommendation in location-based social
networks using user check-in data,” in Proceedings of the 21st ACM SIGSPATIAL International
Conference on Advances in Geographic Information Systems. ACM, 2013, pp. 374–383.
4. J.B. Griesner, T. Abdessalem, and H. Naacke, “Poi recommendation: Towards fused matrix
factorization with geographical and temporal influences,” in Proceedings of the 9th ACM Conference
on Recommender Systems. ACM, 2015, pp. 301–304.
5. C. Cheng, H. Yang, I. King, and M. R. Lyu, “Fused matrix factorization with geographical and social
influence in location-based social networks,” in Twenty-Sixth AAAI Conference on Artificial
Intelligence, 2012.
6. X. Liu, Y. Liu, K. Aberer, and C. Miao, “Personalized point-of-interest recommendation by mining
users’ preference transition,” in Proceedings of the 22nd ACM international conference on
Conference on information & knowledge management. ACM, 2013, pp. 733–738.
7. Q. Yuan, G. Cong, Z. Ma, A. Sun, and N. M. Thalmann, “Time-aware point-of-interest
recommendation,” in Proceedings of the 36th international ACM SIGIR conference on Research and
development in information retrieval. ACM, 2013, pp. 363–372.