Identifying Your Customers in Social Networks Date : 2015/03/12 - - PowerPoint PPT Presentation

Identifying Your Customers in Social Networks

Date : 2015/03/12 Author: Chun-Ta Lu, Hong-Han Shuai, Philip S. Yu Source: ACM CIKM’14 Advisor: Jia-ling Koh Speaker: Han, Wang

1

Outline

Introduction Approach Experiments Conclusion

2

Introduction

Motivation:

Social networks are influential sources in shaping a customer’s attitudes and behaviors. Interactions with friends in social networks of customers are barely observable in most e-commerce companies.

What did Emma Watson buy? Who bought a iPhone 6?

3

Introduction

Purpose:

To fulfill the gap between companies and social networks. Find the one-to-one pair between two networks.

4

Introduction

Challenges:

Difference in network schema. Partially aligned networks.

i Phone 6 PSP i Pad NDS

5

Bipartite Heterogeneous network

Outline

Introduction Approach Experiments Conclusion

6

Approach

Notation:

7

i Phone 6 PSP i Pad NDS

Approach

Problems:

How to extract informative features for the customer identification task using basic information available in most e-commerce sites?

• How to effectively match all the customers, who can be identified in

social networks, to their corresponding social accounts?

8

Approach

Framework:

9

Input Output

Customer-Social Identification (CSI)

gc, gs, A(identified pairs) K(user-specified value) A’ (predicted pairs) 1. Extracting features across networks with different schema 2. Identifying customers in partially aligned networks

Approach

Extract features: (user profile similarity)

Names: (five features)

• Exact username match
• Jaccard similarity
• Distance traveled when typing keyboard
• Longest common sequence
• Levenshtein edit distance

Email Address: For verification of identification.

10

Levenshtein edit distance Ex: Kitten -> Sitting sitten （k→s) sittin （e→i） sitting （→g）

Approach

Extract features: (user interest similarity)

11

i Phone 6 PSP i Pad NDS

1) Common Interest (CI) 2) Jaccard’s Coefficient (JC)

Approach

Extract features: (user interest similarity)

12

i Phone 6 PSP i Pad NDS

3) Admic/Adar Index(AA) 4) Resource Allocation Index (RA)

i Phone 6 i Pad i Phone 6 i Pad

Approach

Extract features: (user interest similarity)

13

i Phone 6 PSP i Pad NDS

3) Katz’s Index: (Set lmax=2)

Approach

Identifying customers:

top-K maximum similarity & stable matching problem

14

threshold = 0.5 ignore candidate pair when K=1

Approach

Identifying customers:

15

• 1. Select the pair of accounts

with the maximum similarity score from candidate pairs.

• 2. If both the users in pair

haven’t assigned to any account, add pair to A’, and set accounts as

• ccupied.
• 3. Doing until finding top K

pairs or no left in A.

Outline

Introduction Approach Experiments Conclusion

16

Experiments

Dataset: (Kickstarter v.s Twitter)

from Nov. 2012 to Sep. 2013 (10 months) Twitter:(Regarding Kickstarter)

• 385K tweets, 178K users, 5.4 million social links, 3725 projects

and 234K links between projects and Twitter users. Kickstarter:

• 3725 projects, 545K customers and 868K adoption links.

17

Experiments

Comparative methods

Unsupervised Link Prediction Supervised Link Prediction Customer-Social Identification (CSI)

• Evaluation Measures: Precision,Recall, F1-measure and ROC

curve(AUC)

18

Experiments

5-fold cross validation unsupervised methods.

19

Experiments

20

Experiments

21

Experiments

22

Outline

Introduction Approach Experiments Conclusion

23

Conclusion

This paper described and studied the problem of customer identification in social networks. Extract two types of features, user profile and user interest, that can be used to compute the similarity scores of pairs between networks. CSI Approach can effectively connect customers and accounts in social network, and outperforms other commonly-used baseline on customer identification.

24