1. L IFE - STAGE P REDICTION FOR P RODUCT R ECOMMENDATION IN E- COMMERCE Speaker: Jim-An Tsai Advisor: Jia-ling Koh Author: Peng Jiang, Yadong Zhu, Yi Zhang, Quan Yuan Date: 2015/9/24 Source: KDD’15 1

2. O UTLINE Introduction Method Experiment Conclusion 2

3. I NTRODUCTION 3

4. 4

5. FRAMEWORK User Predict for User’s life-stage Generate Recommedation One-kid family MESMM Multi-kids family GMM 5

6. O UTLINE Introduction Method Experiment Conclusion 6

7. N OTATIONS 7

8. M AXIMUM E NTROPY S EMI M ARKOV M ODEL (MESMM) 8

9. MESMM 9

10. F EATURES FOR THE C LASSIFIER 1. Category features 10

11. F EATURES FOR THE C LASSIFIER 2. Query A user behavior sequence may contain user product search activities. User search queries can directly reflect users’ requirements, which can indirectly reflect a baby’s age. 3. Product property features Taobao is a distributed market space where sellers sell products. Many sellers provide meta data about the products. 11

12. F EATURES FOR THE C LASSIFIER 4. Product title features Product titles are created by sellers who are not affiliated with Taobao. Sellers are very creative about product titles. As a result, many product titles can be very informative about the life stage of the consumer. 5. Temporal Effect of Features 12

13. F EATURES FOR THE C LASSIFIER 13

14. R ECOMMENDATION B ASED ON L IFE - STAGE To make product recommendations, we propose a model to estimate the probability of a user purchasing a product at a specific age a. P ( p_productj ) be the probability of the user purchasing the product j P ( a|p_productj ) be the conditional probability of making the purchase at age a given that the user will purchase product j. x is a feature vector that represents a purchasing action, w is learnt by maximizing the likelihood of the training data. 14

15. MIXTURE MODEL FOR MULTI-KIDS Gaussian Mixture Model (GMM): 15

16. MIXTURE MODEL FOR MULTI-KIDS (b) Gaussian mixture model with two components for user purchasing age. One component represents 11-month old, and the other one represents 40-month old. 16

17. O UTLINE Introduction Method Experiment Conclusion 17

18. E XPERIMENTAL S ETUP 18

19. L IFE - STAGE P REDICTION AND S EGMENTATION 19

20. L IFE - STAGE P REDICTION AND S EGMENTATION 20

21. A NALYSIS OF T EMPORAL E FFECT 21

22. A NALYSIS OF M ULTI - KIDS 22

23. O NLINE E XPERIMENTS 23

24. O NLINE E XPERIMENTS 24

25. O UTLINE Introduction Method Experiment Conclusion 25

26. C ONCLUSION 1. We introduced the conception of life stages into Ecommerce recommendation systems 2. We proposed a new Maximum Entropy Semi Markov model for stochastic life stage segmentation and prediction 3. We developed a practical efficient large scale industry solution for life stage segmentation and labeling in mum-baby domain, where the life stage transition is deterministic 26

27. C ONCLUSION 4. We proposed a probability model which can explicitly incorporate the life-stage information into an Ecommerce recommender system. 5. We proposed a solution for modeling multi-kids scenario via Gaussian mixture models. 6. We verified of the effectiveness of the life-stage based approach in both offline and online scenarios. 27

28. T HANKS FOR L ISTENING 28