GALLOP: GLOBAL FEATURE FUSED LOCATION PREDICTION FOR DIFFERENT CHECK-IN SCENARIOS
ABSTRACT
Location prediction is widely used to forecast users’ next place to visit based on his/her mobility logs. It is an essentialproblem in location data processing, invaluable for surveillance, business, and personal applications. It is very challenging due to thesparsity issues of check-in data. An often ignored problem in recent studies is the variety across different check-in scenarios, which isbecoming more urgent due to the increasing availability of more location check-in applications.In this paper, we propose a new feature fusion based prediction approach, GALLOP, i.e., GlobAL feature fused LOcation Prediction fordifferent check-in scenarios. Based on the carefully designed feature extraction methods, we utilize a novel combined predictionframework. Specifically, we set out to utilize the density estimation model to profile geographical features, i.e., context information, thefactorization method to extract collaborative information, and a graph structure to extract location transition patterns of users’ temporalcheck-in sequence, i.e., content information. An empirical study on three different check-in datasets demonstrates impressiverobustness and improvement of the proposed approach.
EXISTING SYSTEM:
Behind the check-in data processing, location predic-tion is a fundamental task. However, it is verychallenging due to the check-in data’s inherent characteristics.First, Sparsity: There is a large possible space whichusers can visit, but in fact they only cover a small set ofthe places. Second, Heterogeneousness: Location dataconsists of different kinds of features, i.e., the location, textand temporal information Though some recent work pay attentionto these check-in behavior differences, the inherentvariety between these check-ins and their affect to locationprediction is usually missing. State of the art locationprediction methods can be categorized into three lines, i.e.,collaborative feature based, spatial feature based,and fusion feature based . Collaborative featurebased approaches utilize the Collaborative Filtering or factorizationmethods to cope with the sparsity challenge butthey are sensitive to spatial features. In contrast, spatialfeature based methods use the gravity and locality closenessmeasurements to fit into the location setting, but areusually difficult to generalize the temporal and other relatedfeatures. The feature fusion approach shows the advantageof feature combination to deliver improved accuracy.
PROPOSED SYSTEM:
we propose a new feature fusion approach,i.e., GlobAL feature fusion for LOcation Prediction (GALLOP),to cope with the variety problem in location predic-tion. To improve the applicability of location prediction ap-proach, We utilize several kinds of features and discuss theirdifferent characteristics in the variety of check-in scenarios.Three classes of features are used in GALLOP: contextfeature(geographical aspects), collaboration feature(users’latent interest space) and content feature(places’ descriptionattributes). We introduce intuitive ways to model thesecheck-in features and then formalize a combination frameworkto deliver the predicted target places to end users.The proposed GALLOP prediction approach is not onlygeneral over different check-in scenarios but also compre-hensive of different features. In the context feature, we de-sign a multiple granularity model to profile the geographicalcloseness. We select the predicted candidates based on thecombination of local district, local city and state scales. Theweights of each scale are learned from training dataTo conclude the contribution of this paper, we summarizethe following ones:• First, we investigate the difference of several representativecheck-in scenarios from the spatial andtemporal aspects. We argue that these varying checkinscenarios ask for more general location predictionmethods.• Second, we demonstrate the combination power oflocation features in a novel angle. We not only utilizethe different classes of context, collaboration andcontent information, but also factorize them in anew way to improve the prediction robustness andgenerality.• At last, the extensive study over several real datasetsreveals the improvement and advantage of our approach.We provide an empirical study with severalcompetition methods. Detailed experiments showthe different behaviors of the prediction methods,and prove that the general location prediction approachis a better choice to tackle the location predictionchallenges.
CONCLUSIONS
This paper presents a new feature fusion method for locationprediction problem. We systematically analyze thecheck-in characteristics of different scenarios and proposeto model three categories of features and combine them in aglobal way. The geographical, collaborative and categoricalinformation are all utilized. We propose new models toinclude more global features to improve the generality androbustness of the prediction method. Besides, the approachis versatile and easy to extend. It shows impressive advantageon different datasets and significantly improve theprediction accuracy.This research has several interesting future directions.For example, better ways to improve the feature preprocessingstage and design the compact structure to maintainthe extracted features. It is also valuable to exploitthe evolving factors in the location prediction. Additionally,the feature extraction methods we proposed in this workcan be extended to enable incremental updating. And newcomprehensive location prediction and update setting canbe utilized.
REFERENCES
[1] J. Sang, T. Mei, J.-T. Sun, C. Xu, and S. Li, “Probabilistic sequentialpois recommendation via check-in data,” in Proceedings of the20th International Conference on Advances in Geographic InformationSystems, 2012, pp. 402–405.
[2] X. Liu, Y. Liu, K. Aberer, and C. Miao, “Personalized point-ofinterestrecommendation by mining users’ preference transition,”in Proc. of CIKM, 2013, pp. 733–738.
[3] J.-D. Zhang, C.-Y. Chow, and Y. Li, “Lore: exploiting sequentialinfluence for location recommendations,” in Proceedings of the22nd ACM SIGSPATIAL International Conference on Advances inGeographic Information Systems, 2014, pp. 103–112.
[4] W. Wang, H. Yin, S. Sadiq, L. Chen, M. Xie, and X. Zhou, “SPORE:A sequential personalized spatial item recommender system,” inProc. of ICDE, 2016, pp. 954–965.
[5] D. Lian, X. Xie, V. W. Zheng, N. J. Yuan, F. Zhang, and E. Chen,“CEPR: A collaborative exploration and periodically returningmodel for location prediction,” TIST, vol. 6, no. 1, p. 8, 2015.
[6] A. Noulas, S. Scellato, N. Lathia, and C. Mascolo, “Mining usermobility features for next place prediction in location-based services,”in Proc. of ICDM. IEEE, 2012, pp. 1038–1043.[7] J. Ye, Z. Zhu, and H. Cheng, “Whats your next move: User activityprediction in location-based social networks,” in Proc. of SDM,2013.
[8] A. Noulas, S. Scellato, C. Mascolo, and M. Pontil, “An empiricalstudy of geographic user activity patterns in Foursquare,” in Proc.of ICWSM, 2011, pp. 70–573.
[9] E. Cho, S. A. Myers, and J. Leskovec, “Friendship and mobility:user movement in location-based social networks,” in Proc. ofSIGKDD, 2011, pp. 1082–1090.
[10] J. Bao, Y. Zheng, D. Wilkie, and M. Mokbel, “Recommendations inlocation-based social networks: A survey,” Geoinformatica, vol. 19,no. 3, pp. 525–565, Jul. 2015