Presentation is loading. Please wait.

Presentation is loading. Please wait.

Intelligent DataBase System Lab, NCKU, Taiwan Josh Jia-Ching Ying 1, Eric Hsueh-Chan Lu 2 and Vincent S. Tseng 1 1 Institute of Computer Science and Information.

Similar presentations


Presentation on theme: "Intelligent DataBase System Lab, NCKU, Taiwan Josh Jia-Ching Ying 1, Eric Hsueh-Chan Lu 2 and Vincent S. Tseng 1 1 Institute of Computer Science and Information."— Presentation transcript:

1 Intelligent DataBase System Lab, NCKU, Taiwan Josh Jia-Ching Ying 1, Eric Hsueh-Chan Lu 2 and Vincent S. Tseng 1 1 Institute of Computer Science and Information Engineering National Cheng Kung University No.1, University Road, Tainan City 701, Taiwan (R.O.C.) 2 Department of Computer Science and Information Engineering National Taitung University No.684, Sec. 1, Zhonghua Rd., Taitung City, Taitung County 950, Taiwan (R.O.C.) Followee Recommendation on Asymmetric Location- Based Social Network

2 Intelligent DataBase System Lab, NCKU, Taiwan Outline 2 Introduction Symmetric V.S. Asymmetric Motivation Challenges Geographic-Textual-Social Based Followee Recommendation (GTS-FR) Feature Extraction Model Building Experimental Results Conclusions

3 Intelligent DataBase System Lab, NCKU, Taiwan Symmetric V.S. Asymmetric Types of Social Network Symmetric Social Network Undirected graph Facebook, Gowalla, Foursquare Asymmetric Social Network directed graph Tweeter, Everytrail Friends V.S. Followees 3

4 Intelligent DataBase System Lab, NCKU, Taiwan Introduction – Motivation Most followee recommendations directly adopt concept of traditional friend recommending system. Friend-of-friend link Geographical distance (or similarity) Traditional friend recommending system could not work well for followee recommendation Social feature followee-of-followee Tweeter, Everytrail, … Geographic feature Similarity of users’ trajectory HGSM [Zheng et al., 2011] 4 Trajectory 1 2 3 A hiking… B Shopping… C hiking… Trajectory 1 2 3 1 2 3 A hiking B Shopping… C hiking…

5 Intelligent DataBase System Lab, NCKU, Taiwan Introduction – Motivation 5 Textual Information

6 Intelligent DataBase System Lab, NCKU, Taiwan Introduction – Challenges 6 How to recommend followees for users based on not only social factors but also user-generated data (textual and geographical data)? Social properties Geographical properties Textual properties How to build a model for recommending for followee recommendation?

7 Intelligent DataBase System Lab, NCKU, Taiwan Outline 7 Introduction Symmetric V.S. Asymmetric Motivation Challenges Geographic-Textual-Social Based Followee Recommendation (GTS-FR) Feature Extraction Model Building Experimental Results Conclusions

8 Intelligent DataBase System Lab, NCKU, Taiwan Feature Extraction 8 Social Property (SP) Geographical Property (GP) Textual Property (TP)

9 Intelligent DataBase System Lab, NCKU, Taiwan Social Property (SP) 9 where T(u, v) indicates the set of Transition- Setters of all followee-of-followee links from u to v. j and a are Transition-Setters of followee-of- followee links i to b

10 Intelligent DataBase System Lab, NCKU, Taiwan Social Property – Transitivity 10 Transitivity by Links between Followees and Followers (LinkTran) based on users’ following relation Transitivity by Communications between Followees and Followers (CTran) based on users’ textual information

11 Intelligent DataBase System Lab, NCKU, Taiwan Social Property – Transitivity 11 Transitivity by Links between Followees and Followers (LinkTran ) Proportion of the user’s followers who also follow the his followees The followers of user j are user a user i, and user k. The followee of user j is user b. Thus, the LinkTran of user j is (1+0+0)/(3×1) ≒ 0.33

12 Intelligent DataBase System Lab, NCKU, Taiwan Social Property – Transitivity 12 Transitivity by Communications between Followees and Followers (CTran)

13 Intelligent DataBase System Lab, NCKU, Taiwan Feature extraction 13 Social Property (SP) Geographical Property (GP) Textual Property (TP)

14 Intelligent DataBase System Lab, NCKU, Taiwan Geographical Property (GP) 14 where Tr(u) indicates the set of trajectories of user u. … … uvSimilar?

15 Intelligent DataBase System Lab, NCKU, Taiwan Trajectory Similarity 15 Y. Zheng, L. Zhang, and X. Xie. Recommending friends and locations based on individual location history. ACM Transaction on the Web, 2011.

16 Intelligent DataBase System Lab, NCKU, Taiwan Trajectory Similarity 16 P = Q = their longest common sequence is LCS(P, Q) = where I T (P, Q) is an indicator function which indicates whether the tags of P and Q are the same.

17 Intelligent DataBase System Lab, NCKU, Taiwan Feature extraction 17 Social Property (SP) Geographical Property (GP) Textual Property (TP)

18 Intelligent DataBase System Lab, NCKU, Taiwan Textual Property (GP) 18 w1w1 w2w2 w3w3 w4w4 w5w5 User-Keyword (UK) Graph User u i has texted keyword w j in some textual information for c ij times. HITS-Based random walk model M = [c ij ] c 12 UK-based Textual Property Users comment other users’ trip or write travelogue within their trips can represent their information needs.

19 Intelligent DataBase System Lab, NCKU, Taiwan Textual Property (GP) 19 w1w1 w2w2 w3w3 w4w4 w5w5 l2l2 l3l3 l4l4 l5l5 l6l6 l1l1 l7l7 Location-Keyword (LK) Graph Location l s has been texted with keyword w j in comments or travelogues for v sj times. HITS-Based random walk model N = [v sj ] v sj ULK-based Textual Property similar keywords could be texted with the similar locations.

20 Intelligent DataBase System Lab, NCKU, Taiwan Outline 20 Introduction Symmetric V.S. Asymmetric Motivation Challenges Geographic-Textual-Social Based Followee Recommendation (GTS-FR) Feature Extraction Model Building Experimental Results Conclusions

21 Intelligent DataBase System Lab, NCKU, Taiwan Model Building 21 User IDTextual Properties Geographical Properties Social Properties Follow j0.70.20.5 1 (i  j) …………… m0.10.90.5 0 (i  j) SVM, Logistic Regression, C4.5, … We choose SVM as the classifier because it has shown excellent performance in similar tasks user i

22 Intelligent DataBase System Lab, NCKU, Taiwan Outline 22 Introduction Symmetric V.S. Asymmetric Motivation Challenges Geographic-Textual-Social Based Followee Recommendation (GTS-FR) Feature Extraction Model Building Experimental Results Conclusions

23 Intelligent DataBase System Lab, NCKU, Taiwan Experimental Evaluation 23 Experimental dataset – EveryTrail Dataset We extract the data from 12/2011 to 3/2012, each month is a time period. We got 35,153 users and 4 snapshots. Snapshot1st2nd3rd4th # of trips116179145,662193,331196,949 # of comments337,519293,453315,585379,020 # of links700,103777,7381,056,0771,139,832

24 Intelligent DataBase System Lab, NCKU, Taiwan Experimental Evaluation 24 Experimental measurements p- indicate the number of incorrect recommendations p+ indicate the number of correct recommendations R indicates the total number of links in the testing data

25 Intelligent DataBase System Lab, NCKU, Taiwan Comparison of Various Features 25 The Textual Properties is more effective than other properties for followee recommendation.

26 Intelligent DataBase System Lab, NCKU, Taiwan Comparison of Various Features (cont.) 26 In detail, The ULK-based Textual Property is more effective than other properties for followee recommendation.

27 Intelligent DataBase System Lab, NCKU, Taiwan Comparison with Existing Recommenders 27 To compare Existing Recommenders, our approach significantly outperform other Recommenders.

28 Intelligent DataBase System Lab, NCKU, Taiwan Conclusions We have proposed a novel approach named Geographic- Textual-Social Based Followee Recommendation (GTS-FR) for followee recommendation. We propose three kinds of useful features Social Property (SF), Geographical Property (GP) Textual Property (TP) Through a series of experiments by the real dataset obtained from EverTrail, we have validated our proposed GTS-FR and shown that GTS-FR has excellent effectiveness.

29 Intelligent DataBase System Lab, NCKU, Taiwan Question? Thank you for your attentions


Download ppt "Intelligent DataBase System Lab, NCKU, Taiwan Josh Jia-Ching Ying 1, Eric Hsueh-Chan Lu 2 and Vincent S. Tseng 1 1 Institute of Computer Science and Information."

Similar presentations


Ads by Google