24
Association Rules in Web Usage Logfile Data Empirical Insights into the Use of User-Generated Web Site Features International Conference on Electronic Commerce 2013 Turku, Finland Aug. 13, 2013 Dr. Christian Holsing and Dr. Carsten D. Schultz Chair of Marketing, University of Hagen, Germany Research supported by SAS Institute Germany

Konsumentenverhalten im Social Shopping - empirische Analyse mittels Logfiles

Embed Size (px)

DESCRIPTION

Konsumentenverhalten im Social Shopping - empirische Analyse mittels Logfiles. Methode: Assoziationsregeln.

Citation preview

Page 1: Konsumentenverhalten im Social Shopping - empirische Analyse mittels Logfiles

Association Rules in Web Usage

Logfile Data – Empirical Insights into

the Use of User-Generated

Web Site Features

International Conference on Electronic Commerce 2013

Turku, Finland

Aug. 13, 2013

Dr. Christian Holsing and Dr. Carsten D. Schultz

Chair of Marketing, University of Hagen, Germany

Research supported by

SAS Institute Germany

Page 2: Konsumentenverhalten im Social Shopping - empirische Analyse mittels Logfiles

Overview

2

1. Relevance and Basics of Business Model SSC

2. Literature Review

3. Research Question/Methodology

4. Empirical Results (Logfile Analysis)

5. Conclusion and Outlook

Page 3: Konsumentenverhalten im Social Shopping - empirische Analyse mittels Logfiles

University of Hagen

3

Largest university in German-speaking countries

> 80,000 students

Distance Learning System

50 study centres in Germany, Austria, Switzerland, and Central and Eastern Europe

Faculties:

Cultural and Social Sciences

Mathematics and Computer Science

Business Administration and Economics

Law

www.fernuni-hagen.de/marketing

Page 4: Konsumentenverhalten im Social Shopping - empirische Analyse mittels Logfiles

Relevance of Business Model SSC

4

Web 2.0 provides consumers with many methods of

creating and sharing user-generated content (UGC)

Social media are growing rapidly

Social Networking + Online-Shopping = Social Shopping

Social Shopping is about connecting consumers and

shopping together

Business model Social Shopping Community (SSC)

becomes more relevant

polyvore.com: more than 21 Mio. Unique Visitors/Month;

22 Mio. $ Venture Capital

Page 5: Konsumentenverhalten im Social Shopping - empirische Analyse mittels Logfiles

SSC: Definition

5

OLBRICH/HOLSING 2011, p. 15:

A SSC is an online-shopping service that connects

consumers and lets them discover, share, recommend,

rate, and purchase products.

In contrast to traditional e-commerce channels, such as

online-shops, and shopbots, SSCs additionally offer user-

generated social-shopping features, as well as potential

interaction, so as to initiate or simplify purchase decisions.

Page 6: Konsumentenverhalten im Social Shopping - empirische Analyse mittels Logfiles

SSC: Features

6

Product Detail-Site

at smatch.com:

Page 7: Konsumentenverhalten im Social Shopping - empirische Analyse mittels Logfiles

SSC: Example of a Style on polyvore.com

7

Page 8: Konsumentenverhalten im Social Shopping - empirische Analyse mittels Logfiles

Literature Review: Social Shopping

Research in Social Shopping is just at the beginning

Only few aspects are analyzed, e.g., impact of user-

generated content on economic outcomes

(GODES/MAYZLIN 2004; CHEVALIER/MAYZLIN 2006; LIU

2006; MOE/TRUSOV 2011)

Some recent studies are analyzing Social Shopping/

SSCs more detailed:

KANG/PARK 2009: Acceptance Factors of Social Shopping

SHEN/EDER 2011: An Examination of Factors Associated

with User Acceptance of Social Shopping Websites

8

Page 9: Konsumentenverhalten im Social Shopping - empirische Analyse mittels Logfiles

Literature Review: Logfile Analysis/E-Commerce

9

Authors Country, Data Focus Sessions

BUCKLIN/SISMEIRO 2003 n. a., 10/1999 Car website 6,630 sessions

HUANG/LURIE/MITRA 2009 USA, 01 - 07/2004 comScore panel: websites in 6 product categories 210 sessions

JOHNSON/MOE/FADER/BELLMAN/LOHSE 2004 USA, 07/1997 - 06/1998 Media Metrix panel: 51 websites (books, CDs, flights) 33,452 unique visits

MOE 2003 n. a., 5/18 - 7/05/2000 Online shop for nutrition products 5,730 users; 7,143 sessions

MONTGOMERY/LI/SRINIVAN/LIECHTY 2004 USA, 4/01 - 4/30/2002 Media Metrix panel: barnesnoble.com, books.com,

bn.com

1,160 users; 1,659 sessions

PARK/CHUNG 2009 USA, 07 - 12/2004 comScore panel: travel websites (Expedia, etc.) Sessions of 1,190 panelists

PARK/FADER 2004 USA, 10/1997 - 05/1998 Media Metrix panel: online shops for books, and CDs 7,377 panelists; 18,027

sessions

VAN DEN POEL/BUCKINX 2005 n. a., 5/25 - 4/18/2002 Online shop for wine 1,382 visitors; 10,173

sessions

ZHANG/FANG/SHENG 2006 USA, 07 - 12/2002 comScore panel: 69 websites (CDs, computer hardware,

flight tickets)

104,416 sessions

This study Germany, Austria,

Switzerland, 5/01 -

10/31/2009

SSC focussing on fashion, living, and lifestyle 2.9 million sessions

Page 10: Konsumentenverhalten im Social Shopping - empirische Analyse mittels Logfiles

Literature Review: Clickstream Studies

Clickstream data are a powerful source of information

Using clickstream data confronts researchers with a number of

difficulties, e.g.:

Capturing the purchasing environment of consumers

Associated data pre-processing

Accordingly, relatively few studies in fact use such data

PADMANABHAN/ZHENG/KIMBROUGH 2001; MOE/FADER 2004;

SISMEIRO/BUCKLIN 2004; VAN DEN POEL/BUCKINX 2005, PARK/CHUNG 2009

Research gap:

Analyzing consumer behavior in SSC‘s

Analyzing impact of more than just one kind of user-generated content,

e.g., ratings

Focus on categories of fashion, living, and lifestyle

10

Page 11: Konsumentenverhalten im Social Shopping - empirische Analyse mittels Logfiles

Research Question/Methodology

Which shopping features, especially user-

generated features, of a SSC are used

together within user sessions?

Data: Web usage logfiles of a SSC

Method: Association Rule Learning

we will identify strong rules, and thus structural

relations between user-generated and direct

shopping features

using different measures of interestingness

11

Page 12: Konsumentenverhalten im Social Shopping - empirische Analyse mittels Logfiles

Logfile Analysis: Data and Process

12

Logfiles of a high-traffic SSC

Categories of fashion, living, and lifestyle

> 600 participating online shops

Product data base > 1.5 million products

Period from May 1st, 2009 to October 31st, 2009

Number of sessions: 2.9 million

4 variable categories: general, direct shopping,

social shopping, and transactional

Software: SAS Enterprise Miner 6.2

Page 13: Konsumentenverhalten im Social Shopping - empirische Analyse mittels Logfiles

Variables (4 Categories)

13

General Home (number of home page visits)

Product (number of product-detail sites visited)

Direct-Shopping Filter mechansims (brand, category, gender, price, sale, shop)

Search field

Social-Shopping (user-generated Web site features) List

Style

Profile

Tag

Transactional Click out (number of visits to participating online shops)

Page 14: Konsumentenverhalten im Social Shopping - empirische Analyse mittels Logfiles

Descriptive Statistics

14

Variable Min Max Mean SD

General:

HOME 0 130 .09 .450

PRODUCT 0 664 .91 2.032

Direct-Shopping:

SEARCH_BRAND 0 369 .31 2.492

SEARCH_CAT 0 557 1.48 6.669

SEARCH_FIELD 0 520 1.15 2.548

SEARCH_GENDER 0 430 .73 4.016

SEARCH_PRICE 0 220 .12 1.693

SEARCH_SALES 0 234 .05 .960

SEARCH_SHOP 0 178 .12 .905

Social Shopping:

LIST 0 112 .02 .227

STYLE 0 95 .01 .164

PROFILE 0 72 .01 .148

TAG 0 183 .03 .565

Transactional:

CLICK_OUTS 0 471 .81 1.878

Page 15: Konsumentenverhalten im Social Shopping - empirische Analyse mittels Logfiles

Method of Association Rules Learning

15

Set of user sessions S = {s1, s2, …, sn}

A user session is a sequence of interactions

I = {i1, i2, …, im}

Association rule is

an implication of A B

where A, B I and A B = Ø

{HOME, PRODUCT} {CLICK_OUT}

Page 16: Konsumentenverhalten im Social Shopping - empirische Analyse mittels Logfiles

Measures of Association Rules

16

Significance measure

Quality measure

Interestingness measure

S

sBASsBA

})(|{)sup(

})(|{

})(|{)(

sASs

sBASsBAconf

)sup(

)()(

B

BAconfBAlift

Page 17: Konsumentenverhalten im Social Shopping - empirische Analyse mittels Logfiles

Summary of Association Rules

17

Conclusion min.

support

min.

confident

max.

antecedents

number of

assoc. rules

CLICK_OUT .01 .05 3 32

PRODUCT .01 .05 3 34

LIST .007 .03 3 3

PROFILE .007 .03 3 3

STYLE .007 .03 3 4

TAG .01 .05 3 19

Page 18: Konsumentenverhalten im Social Shopping - empirische Analyse mittels Logfiles

Results of Association Rules Learning (1)

18

Conclusion Antecedent No. sup conf lift

{CLICK_OUT} {HOME, PRODUCT} 84,103 .0289 .5812 1.41

{CLICK_OUT} {PRODUCT, SEARCH_GENDER} 115,478 .0397 .5423 1.32

{CLICK_OUT} {PRODUCT, SEARCH_GENDER,

SEARCH_CAT} 61,287 .0211 .5284 1.28

{PRODUCT} {HOME, CLICK_OUT} 59,140 .0200 .8407 1.94

{PRODUCT} {TAG} 31,722 .0110 .7654 1.77

{PRODUCT} {SEARCH_CAT,

SEARCH_GENDER, CLICK_OUT} 41,702 .0143 .7238 1.67

Page 19: Konsumentenverhalten im Social Shopping - empirische Analyse mittels Logfiles

Results of Association Rules Learning (2)

19

Conclusion Antecedent No. sup conf lift

{LIST} {STYLE} 23,842 .0082 .0927 8.31

{LIST} {HOME, PRODUCT, SEARCH_FIELD} 24,530 .0084 .0456 4.09

{LIST} {HOME, PRODUCT, SEARCH_CAT} 26,548 .0091 .0326 2.92

{PROFILE} {STYLE} 23,842 .0082 .1088 28.01

{PROFILE} {LIST} 32,486 .0112 .0783 20.17

{PROFILE} {PRODUCT, LIST} 22,198 .0076 .0692 7.81

{STYLE} {PRODUCT, LIST} 22,198 .0076 .0711 8.69

{STYLE} {LIST} 32,486 .0112 .0680 8.31

{STYLE} {HOME, PRODUCT, SEARCH_FIELD} 24,530 .0084 .0419 5.11

{TAG} {PRODUCT, SEARCH_BRAND} 47,696 .0165 .3275 30.05

{TAG} {SEARCH_BRAND, SEARCH_CAT} 46,771 .0161 .2700 24.78

{TAG} {SEARCH_BRAND, SEARCH_FIELD,

CLICK_OUT} 29,553 .0102 .1709 15.68

Page 20: Konsumentenverhalten im Social Shopping - empirische Analyse mittels Logfiles

Implic@tions

20

Association rules provide insights into structural relationships in user sessions

recommendations can be derived to improve the use and usability, e.g., linking certain shopping features

Identifying features that support main economic aim: click-out Social shopping features: no strong relationships with click-out

Potential strategy: adjust features, e.g., by integrating a direct click-out into styles and lists, instead of having product-detail sites as an intermediate step

Social shopping features: highly associated to each other Way of increasing click-outs: loosen the linkage between these features

However, one important user motive may be to browse and participate in the community manage specific user groups

Page 21: Konsumentenverhalten im Social Shopping - empirische Analyse mittels Logfiles

Implic@tions

21

Provide different features to various user types e.g., to community-orientated users, browers, buyers, etc.

specific cluster analysis or self-organizing maps (SOM)

Split testing could evaluate such a solution before implementation

Provide sales promotions within lists, profiles, and styles increase click-out rate

Search results may also include direct links to online shops e.g., by miniature previews, in addition to product-detail sites

Management needs to monitor consumer confusion or reactance

Overall, association rules provide evidence enabling the management to reduce user navigation and search effort increase usability

Page 22: Konsumentenverhalten im Social Shopping - empirische Analyse mittels Logfiles

Limitations and Future Research

22

Future research should confirm results and extend the focus to other features and to different types of online services

As user-generated features continue to evolve dynamically, more recent data can incorporate the latest developments

Method of Association Rules Learning

does not consider the order of interactions within a session

Rules simply consider request for an interaction, not frequency

good starting point to identify interesting relations

further inspection: order (clickstream) and frequency of interactions

Distinguish between different user groups to analyze potential differences between these segments

Page 23: Konsumentenverhalten im Social Shopping - empirische Analyse mittels Logfiles

Conclusion and Outlook!

23

We enhance the research in Social Shopping

It seems likely that Social Shopping will become

more and more important

Use of social media increases

New business models arise, e.g., Pinterest (online

pinboard)

New technologies will be established rapidly (mobile,

tablets, etc.)

Booz&Co forecast: social commerce revenues will hit

$30bn by 2015

Page 24: Konsumentenverhalten im Social Shopping - empirische Analyse mittels Logfiles

Thank You

For Your Attention!

Dr. Christian Holsing and Dr. Carsten D. Schultz

Contact:

Dr. Christian Holsing: http://social-commerce.net, www.lynx-ecommerce.de

Dr. Carsten Schultz: www.fernuni-hagen.de/marketing