Spatio-Temporal Topic Modeling in Mobile Social Media for Location Recommendation
ICDM, pp. 1073-1078, 2013.
EI
Weibo:
Abstract:
Mobile networks enable users to post on social media services (e.g., Twitter) from anywhere and anytime. This new phenomenon led to the emergence of a new line of work of mining the behavior of mobile users taking into account the spatio-temporal aspects of their engagement with online social media.
In this paper, we address the problem o...More
Code:
Data:
Introduction
- Location recommender systems can recommend a set of locations that users may be interested in, based on the history of user check-ins.
- Geographical influence suggests that locations that are closer to the user’s visited locations are recommended with higher probabilities, and the existing methods [15], [9] assume that the location recommendations to a user should be geographically regularized by the set of all of the user’s check-ins.
- A good recommendation should be one of those locations in either one of the regions, but the existing models will recommend locations along the commute route, since they are on average closer to the user’s check-ins in both regions
Highlights
- Location recommender systems can recommend a set of locations that users may be interested in, based on the history of user check-ins
- Geographical influence suggests that locations that are closer to the user’s visited locations are recommended with higher probabilities, and the existing methods [15], [9] assume that the location recommendations to a user should be geographically regularized by the set of all of the user’s check-ins
- We propose the first spatio-temporal topic model for location recommendation, capturing the geographical influence between user regions and locations, and temporal activity patterns of different topics and locations
- We demonstrate that the proposed SpatioTemporal Topic model consistently improves the test perplexity and the average accuracy@1,5,10 for location recommendation, compared to existing state-of-the-art recommendation algorithms and geographical and temporal topic models
- We propose the first spatio-temporal topic model to capture the spatial and temporal aspects of check-ins, as well as user profiles in a single probabilistic model, called Spatio-Temporal Topic (STT) model
- Our experiments demonstrate substantially improved performance in location recommendation
Methods
- #. Avg. check-ins/user Avg. check-ins/location Twitter Gowalla Brightkite.
- The authors focus on the tasks of location recommendation for users.
- Given a check-in with a user, the task is to recommend top-k locations, that user will visit in the future.
- Given the user u and time t of a check-in d, the probability that user u visits location i at time t is computed by p(i|t, u, Θ) ∝
Results
- Evaluation Metrics
Perplexity is the standard for measuring how well a probabilistic model fits the data, and is monotonically decreasing in the likelihood of the test data set, so that a lower perplexity indicates better performance of the model. - The top-k accuracy for a test check-in is one when the ground truth location is in the top-k recommendations, and zero otherwise.
- The authors evaluate the following comparison partners, which all model either the coordinates or the index of locations:.
- M R (Multi-Region)
- This is a simplified version of the STT model.
- It assumes that users are associated with region distributions, and the coordinates of locations are drawn from 2D Gaussian distributions.
Conclusion
- The authors propose the first spatio-temporal topic model to capture the spatial and temporal aspects of check-ins, as well as user profiles in a single probabilistic model, called Spatio-Temporal Topic (STT) model.
- STT exploits the interdependencies between users’ regions and their locations, and between temporal activity patterns and locations.
- The authors compare STT against state-of-the-art methods in the areas of recommender systems, and geographical and temporal topic modeling.
- The authors' experiments demonstrate substantially improved performance in location recommendation
Summary
Introduction:
Location recommender systems can recommend a set of locations that users may be interested in, based on the history of user check-ins.- Geographical influence suggests that locations that are closer to the user’s visited locations are recommended with higher probabilities, and the existing methods [15], [9] assume that the location recommendations to a user should be geographically regularized by the set of all of the user’s check-ins.
- A good recommendation should be one of those locations in either one of the regions, but the existing models will recommend locations along the commute route, since they are on average closer to the user’s check-ins in both regions
Methods:
#. Avg. check-ins/user Avg. check-ins/location Twitter Gowalla Brightkite.- The authors focus on the tasks of location recommendation for users.
- Given a check-in with a user, the task is to recommend top-k locations, that user will visit in the future.
- Given the user u and time t of a check-in d, the probability that user u visits location i at time t is computed by p(i|t, u, Θ) ∝
Results:
Evaluation Metrics
Perplexity is the standard for measuring how well a probabilistic model fits the data, and is monotonically decreasing in the likelihood of the test data set, so that a lower perplexity indicates better performance of the model.- The top-k accuracy for a test check-in is one when the ground truth location is in the top-k recommendations, and zero otherwise.
- The authors evaluate the following comparison partners, which all model either the coordinates or the index of locations:.
- M R (Multi-Region)
- This is a simplified version of the STT model.
- It assumes that users are associated with region distributions, and the coordinates of locations are drawn from 2D Gaussian distributions.
Conclusion:
The authors propose the first spatio-temporal topic model to capture the spatial and temporal aspects of check-ins, as well as user profiles in a single probabilistic model, called Spatio-Temporal Topic (STT) model.- STT exploits the interdependencies between users’ regions and their locations, and between temporal activity patterns and locations.
- The authors compare STT against state-of-the-art methods in the areas of recommender systems, and geographical and temporal topic modeling.
- The authors' experiments demonstrate substantially improved performance in location recommendation
Tables
- Table1: NOTATIONS OF PARAMETERS
- Table2: STATISTICS OF DATA SETS FROM TWITTER, GOWALLA, AND BRIGHTKITE
Related work
- Location Recommendation. Many traditional recommendation algorithms, such as MF (Matrix Factorization) [8], [11], can be employed for location recommendation. LDA (Latent Dirichlet Allocation) [1] can also be used for location recommendation. However, none of them considers the spatiotemporal aspects for location recommendation.
Recent works [14], [15], [17], [2], [9] focus on capturing the geographical influence for location recommendation. [15], [14], [17] extend CF, [2] extends MF, and [9] proposes GLDA (Geo Latent Dirichlet Allocation) that extends LDA. Although [14], [17], [15], [2] capture the geographical influence, and the user and item factors (interests), they fuse these two parts in a linear regression framework with user input weights.
Reference
- D. M. Blei, A. Ng, and M. Jordan. Latent dirichlet allocation. JMLR, 3:993–1022, 2003.
- C. Cheng, H. Yang, I. King, and M. R. Lyu. Fused matrix factorization with geographical and social influence in location-based social networks. AAAI, 2012.
- Z. Cheng, J. Caverlee, K. Lee, and D. Z. Sui. Exploring millions of footprints in location sharing services. ICWSM, 2011.
- E. Cho, S. A. Myers, and J. Leskovec. Friendship and mobility: user movement in location-based social networks. KDD, pages 1082–1090, 2011.
- J. Eisenstein, A. Ahmed, and E. P. Xing. Sparse additive generative models of text. ICML, pages 1041–1048, 2011.
- L. Hong, A. Ahmed, S. Gurumurthy, A. J. Smola, and K. Tsioutsiouliklis. Discovering geographical topics in the twitter stream. WWW, pages 769–778, 2012.
- N. Kawamae. Trend analysis model: trend consists of temporal words, topics, and timestamps. WSDM, pages 317–326, 2011.
- Y. Koren, R. Bell, and C. Volinsky. Matrix factorization techniques for recommender systems. Computer, 42:30–37, 2009.
- T. Kurashima, T. Iwata, T. Hoshide, N. Takaya, and K. Fujimura. Geo topic model: joint modeling of user’s activity area and interests for location recommendation. WSDM, pages 375–384, 2013.
- A. Noulas, S. Scellato, N. Lathia, and C. Mascolo. Mining user mobility features for next place prediction in location-based services. In ICDM, pages 1038–1043, 2012.
- R. Salakhutdinov and M. Andriy. Probabilistic matrix factorization. NIPS, pages 1257–1264, 2008.
- M. Schmidt, A. Niculescu-Mizil, and K. Murphy. Learning graphical model structure using l1-regularization paths. AAAI, pages 1278–1283, 2007.
- X. Wang and A. McCallum. Topics over time: a non-markov continuoustime model of topical trends. KDD, pages 424–433, 2006.
- M. Ye, P. Yin, and W.-C. Lee. Location recommendation for locationbased social networks. GIS, pages 458–461, 2010.
- M. Ye, P. Yin, W.-C. Lee, and D.-L. Lee. Exploiting geographical influence for collaborative point-of-interest recommendation. SIGIR, pages 325–334, 2011.
- Z. Yin, L. Cao, J. Han, C. Zhai, and T. Huang. Geographical topic discovery and comparison. WWW, pages 247–256, 2011.
- V. W. Zheng, Y. Zheng, X. Xie, and Q. Yang. Collaborative location and activity recommendations with gps history data. WWW, pages 1029– 1038, 2010.
Full Text
Tags
Comments