Stephen Guo, Mengqiu Wang & Jure Leskovec Stanford University

1. The role of social networks in online shopping:
Information passing, price of trust, and
consumer choice. ACM EC 2011 San Jose,
CA, USA.
Stephen Guo, Mengqiu Wang & Jure Leskovec
Stanford University
Fredrick Awuor

2. Outline
 Introduction
 Paper Objectives
 Case Study: Taobao Networks
 Information Passing
 Price of Trust
 Consumer Choice Prediction
 Open Research Issues
31-Mar-15
2

3. Introduction
 We speak beforehand to the shopkeeper about suitable products we want to
purchase.
 We consult our friends and family before buying something unfamiliar
 We have an urge to tell friends about a popular new product we just bought
 E-commerce websites use recommendation and product comparison to help
customers discover new products.
 Can we substitute this (anonymous opinions) with personalized recommendations one
receives from a friend or relative?
31-Mar-15
3

4. Paper Objective
 How do friends influence consumer purchasing decisions and product adoption?
 What factors influence the success of word of-mouth product recommendations?
 How does social influence and reputation affect commercial activity?
Contribution: Establish relationship between social and e-commercial network
Insight: Information passing, price of trust, consumer choice prediction
 Information passing: an individual purchases a product, then messages a friend, what is
the likelihood that the friend will then purchase the product? Where will he purchase it
from?
31-Mar-15
4

5. Case Study: Taobao Networks
31-Mar-15
5
Chinese consumer marketplace, World largest e-commerce website, (> 380M users 2010).

6. Case Study: Taobao Networks
 Why Taobao? It integrates instant messaging tool which buyers can use to
ask sellers about products or ask other buyers for advice
 Supports B2C, B2B, and C2C
 Dataset describing behavior of 1M users and transactions across over
10,000 products
 User information: transactions with other users (product ID, price, quantity,
timestamp), contact lists, timestamp of messages exchanged (not the
content)
31-Mar-15
6

7. Taobao Networks
 Dyadic buyer-seller relationships:
 Is message activity correlated with
trading activity?
 Positive increasing relationship
between message volume and trade
volume
31-Mar-15
7

8. Taobao Networks
 “Do buyers talk to sellers more about expensive products?”
 Expectedly YES … but how much more are they talked about?
 Count the number of messages sent from buyer to seller on transaction date, assuming
that at least one message is exchanged
31-Mar-15
8
• The number of messages sent is
relatively constant for products of price
below 100 CNY, then increases
logarithmically for products of higher
price.
• Why? Buyers minimize transaction risks
by messaging

9. Taobao Networks
31-Mar-15
9
 To minimize transaction risks buyers speak
with sellers often before transaction to
inquire about product details.
 How often do buyers speak to sellers
before and after trades?
 Most messages occur on the day of
transaction, likely being product
negotiation.
 Post-trade messages are significantly
more common than pre-trade messages

10. Information Passing
 In the Taobao system, buyers have an option of using an escrow service, where
the seller first ships the product, and payment is exchanged after the buyer
examines the product.
 The observed post-trade messages are likely discussion regarding product
satisfaction and payment confirmation.
 Messaging activity is correlated with trading activity across pairs of users
31-Mar-15
10

11. Presence of Information Passing
 Consider: a buyer notices a deal offered at an
electronic store, makes a purchase, then messages his
friend about the deal.
 Will the friend also make a purchase from the same
store?
 How large is the influence of the buyer?
31-Mar-15
11

12. Presence of Information Passing
 The more mutual contacts a pair of users has,
the greater the likelihood that they engaged
in a commercial transaction (Standard)
 For users who have exchanged at least one
message (Msg Req), for a given number of
mutual contacts, the transaction probability is
higher than standard.
31-Mar-15
12

13. Presence of Information Passing
 Inference: Trades are more likely to be embedded in the dense sub-graphs of
communication networks.
 Implication: Social proximity and trade likelihood are correlated
 Information passing and product recommendation present in the Taobao network
 Social proximity: Measured by the number of mutual contacts between the pair
 Conclusion: Messaging increases purchasing behavior in the Taobao network.
31-Mar-15
13

14. Presence of Information Passing
31-Mar-15
14
 Information passing success rate
decreases with product price for the
price range from 1 CNY to 15 CNY, then
increases slightly for products priced
above 15 CNY.
 The large closure probability at a price
of 1 CNY is due to the popularity and
virality of virtual goods, such as game
credits.

15. Price of Trust
 Spread of product recommendations, through word-of-mouth, inherently relies upon
a notion of buyer-buyer trust.
 In the context of electronic marketplaces, buyers are unsure about seller
trustworthiness, so buyers put their trust into seller ratings and reviews, and are willing
to pay a premium to sellers with good reputations
 How much extra will a buyer pay for transaction with a highly rated seller?
31-Mar-15
15

16. Price of Trust
31-Mar-15
16
 Use seller rating as indicator for reputations and trustworthiness
 Use rating information to compare sellers of the same product and determine
how their sale prices differ.
 Higher rating is associated with a seller selling his products at a premium
compared to most of his peers.

17. Price of Trust
31-Mar-15
17
 Observation: seller rating of 97.1%
corresponds to transaction at the median
cluster price.
 Also, buyers are willing to pay more to
highly rated sellers to minimize transaction
risk,
sellers who maintain good
reputations are financially rewarded.

18. Consumer Choice Prediction
31-Mar-15
18
 “How does an online consumer decide upon a seller to purchase from when
there are many sellers offering the same product?”
 Model consumer choice as a machine learning task (utilizing primarily social
networking features)
 Reason: Social graph is a far better determinant of consumer behavior than
metadata features such as seller reputation or product price.

19. Consumer Choice Prediction
 When faced with a selection of substitute goods offered by different sellers, buyers
will not just choose their preferred seller through simple heuristics regarding price or
rating.
 Buyers utilize many sources of information (seller history, advice of friends, seller’s
messages), and each buyer processes the information in their own way in order to
make a personal purchasing decision
 Contribution:
 Develop a machine learning model to predict consumer choice, and
 Show that the social network is the most important feature in predicting how
consumers choose their transaction partners.
31-Mar-15
19

20. Consumer Choice Prediction
 58,812 sets of training examples, (75% train and 25% test sets). The SVMs trained with
linear kernels.
 Evaluation metrics:
 Precision at Top 1 (P@1)- Fraction of times that the top ranked seller is actually the
true seller. (Higher is better)
 Mean Rank (MR)- Average Rank of the true seller. (Lower is better)
 Mean Reciprocal Rank (MRR) - Average Reciprocal Rank of the true seller. (Higher is
better)
 Three rule-based baselines: Random, MinPrice, MostMsg
31-Mar-15
20

21. Consumer Choice Prediction
31-Mar-15
21

22. Open Research Issues
 Analysis of user browsing data to develop refined consumer choice models for
social commerce,
 Study of information passing while factoring in both buyer-buyer and buyer-
seller trust relationships, and
 Viral marketing to influence consumer choice in social commerce.
 Cold start problem – New sellers/products with no ratings
31-Mar-15
22

23. Comments
31-Mar-15
23