Mining Shopping Patterns for Divergent Urban Regions by - PowerPoint PPT Presentation

Mining Shopping Patterns for Divergent Urban Regions by Incorporating Mobility Data Author : Haishan Liu,Tianran Hu, Ruihua Song, Yingzi Wang, Xing Xie, Jiebo Luo Source : CIKM’ 16 Advisor : Jia-Ling Koh Speaker : Chia-Yi Huang Date : 2017/08/29

Outline ▸ Introduction ▸ Method ▸ Experiment ▸ Conclusion 2

Introduction ▸ Motivation 3

Introduction ▸ Shopping Pattern : <Tables, Photography, Digital accessories,…> ▸ Mobility Pattern : <School Dormitories, School Libraries,…> ▸ Region 4

Introduction ▸ Market Basket Analysis ▸ Consumers usually have demands for a group of products ▸ People’s demands are highly related to their lives 5

Outline ▸ Introduction ▸ Method ▸ Experiment ▸ Conclusion 6

Method 7

Method 8

Method ▸ Shopping Patterns Extraction ▸ Browsing log of 
 shopping website ▸ NMF ▸ Ps ▸ Coefficient Matrix ▸ Rs : sum up the weight of location 
 in the same region. 9

Method ▸ Mobility Patterns Extraction ▸ User ID, POI category, 
 latitude, longitude ▸ NMF ▸ Pm ▸ Coefficient Matrix ▸ Rm : sum up the weight 
 of user in the same region. 10

Method ▸ Collective Matrix Factorization 11

Method ▸ Collective Matrix Factorization ▸ d 12

Method ▸ City-wide Interaction Regularization ▸ Gravity Model ▸ City-wide Interaction Regularization 13

Method ▸ Gravity Model ▸ O i , the number of individuals leaving region i ▸ D j , the number of individuals arriving at region j ▸ The distance between two regions, 14

Method ▸ City-wide Interaction Regularization ▸ The more interactions between two regions, the more alike their lifestyles are. 15

Method ▸ Hybrid Model ▸ Combine the collective matrix factorization and interaction regularization 16

Outline ▸ Introduction ▸ Method ▸ Experiment ▸ Conclusion 17

Experiment ▸ Data Set ▸ Online browsing dataset : 250 product categories ▸ Check-in dataset : 1.5 million check-in data, 200 POI categories ▸ Bus dataset : 3 million bus-trip records ▸ Taxi dataset : 1.9 million taxi-trip records 18

Experiment ▸ Baseline ▸ Matrix Factorization (MF) ▸ Collective Matrix Factorization (CMF) ▸ CMF with neighboring information 19


 Experiment ▸ Evaluation ▸ 
 20

Experiment 21

Experiment 22

Experiment 23

Experiment 24

Experiment 25

Outline ▸ Introduction ▸ Method ▸ Experiment ▸ Conclusion 26

Conclusion ▸ Connecting the shopping patterns with the mobility patterns in a region. ▸ Modeling the interactions between regions, and leverage the information of known regions to infer the shopping patterns in unknown regions. 27

THANK YOU