1. Urban Point-of-Interest Recommendation by Mining User Check-in Behaviors 游晟佑 2012.12.5

2. Outline 1.Authors 2.Introduction 3.UPOI Mine Algorithm 4.Experimental Results and Discussions 2

3. Authors Josh Jia-Ching Ying, Eric Hsueh-Chan Lu, Wen-Ning Kuo and Vincent S. Tseng Institute of Computer Science and Information Engineering National Cheng Kung University 3

4. Introduction (1/2) Why use UPOI Mine? – a number of social based recommendation techniques have been proposed in the literature, most of their concepts are only based on the individual or friends' check-in behaviors. (his / her historical data or limited in geographical area) – regression-tree-based predictor, 1 st time use in this kind of research (They asserted) – a real dataset from Gowalla! 4

5. Introduction (2/2) What makes it different? – More comprehensive Steps i) Social Factor, ii) Individual Preference, and iii) POI Popularity for model building For extracting features in i), ii) iii), and feed it into regression tree model -> relevance score -> POI recommendation 5

6. UPOI Mine Algorithm(1/11) 1. Social Factor (SF), ( 朋友在哪邊打卡了 ? 打卡次數 ? 打卡地點 是否與該 user 接近 ?) CheckSim, DisSim 2. Individual Preference (IP) Descriptive features and semantic tags from user check-in POIs Cpref, Hpref 3. POI Popularity (PP) We employ the popularity of POI to make a "maximum likelihood estimation" of the relative between user and POI RP(relative popularity of POI) 把以上三樣 features 的來源餵進 regression tree model 6

7. UPOI Mine Algorithm(2/11) 7

8. UPOI Mine Algorithm(3/11) Features from Social Factor(SF) – given a friend f and a set of POI P, the f’s relative check-ins of a POI p is formulated as: – given a user-POI pair (u, p), the features extracted form Social Factor could be generally formulated as: 8

9. UPOI Mine Algorithm(4/11) – Similarity Measurement In a LBSN data, the most important information is user’s common check-ins and distance among users for user similarity measurement. – Similarity by Common Check-ins (CheckSim) - We employ the χ 2 test for testing relation of check-in behaviors of Gowalla users and their friends. 9

10. UPOI Mine Algorithm(5/11) 10

11. UPOI Mine Algorithm(6/11) – Similarity by Relative Distance (DisSim) where Distance() indicates the Euclidean distance of two base-points and F(u) indicates the set of user u’s friends. 11

12. UPOI Mine Algorithm(7/11) Features from Individual Preference(IP) – In Gowalla website, there are two kinds of semantic tags, i.e., category and highlight – where count(t, p) indicates the number of times the tag t is annotated on the POI p,and T(p) indicates the set of tags of POI p. – the possibility of that a tag ’coffee’ is annotated on a POI is 2 / (2 + 10 + 88) = 0.22 12

13. UPOI Mine Algorithm(8/11) – Accordingly, given a user-POI pair (u, p), the features extracted form Individual Preference could be generally formulated as: 13

14. UPOI Mine Algorithm(9/11) – Cpref(preference of category) (note: 1,0,2,5,0 for user i) The user i’s personal preference of a category tag A is: (1+0+0) / (1+0+2+5+0) = 0.125 – Hpref(preference in Highlight) The user i’s personal preference of a highlight tag a is: (1+2+5) / { (1+2+5) + (1+0)+(0+5)+(0+2) +(0) } = 0.5 14

15. UPOI Mine Algorithm(10/11) Features from POI Popularity(PP) {3, 12, 3, 7, 5} the set of POIs with category tag A are p1, p2, and p5. The total check-in of POI p1, p2, and p5 are 3, 12, and 5, respectively. Thus, the popularity of POI p1 is 3 / (3 + 12 + 5) = 0.15 15

16. UPOI Mine Algorithm (11/11) POI recommendation – We choose M5Prime as the relevance score predictor because it has shown excellent performance in similar tasks 16

17. Experimental Results and Discussions(1/3) Normalized Discounted Cumulative Gain (NDCG) to measure the list of recommended POIs. – NDCG 1.0 means the effectiveness of recommender is pretty good Mean Absolute Error (MAE) to measure the list of recommended POIs as Equation – The lower MAE is, the fewer error is 17

18. Experimental Results and Discussions(2/3) All these means this kind of data is good. Also they compared earlier works to proof this method is good. 18

19. Experimental Results and Discussions(3/3) We compare the performance of UPOI-Mine with TrustWalker [5] and CF-based model [14] in terms of NDCG and MAE 19

20. Thanks for your listening …