Applications of web mining for marketing of online bookstores

1
Applications of web mining for marketing of
online bookstores

• Source Expert systems with applications
  36(2009) 11249-11256
• Authors I-Cheng Yeh, Che-hui Lien, Tao-Ming
  Ting, Chin-Hao Liu
• Presented by Pei-Chun, Tsai
• Date 2009/12/9

2
Outline

• Introduction
• Methodology
• Data collection
• Data process
• Result
• Model 1 4
• Conclusion

3
Introduction

• The exponential growth of the Internet, online
  bookstore not only provides the convenience for
  reading, but also customize individual service,
  both combined to satisfy readers demand of
  knowledge.
• One-to-one marketing.
• Data mining can be used to achieve it.

4
Method (1/3)

• Web mining
• Difference with data mining
• Use name or terminology to search and to collect
  data.
• Three types of web mining
• Web usage mining
• Web structure mining or Web text mining
• Web content mining

5
Method (2/3)

• Association analysis
• It is widely used for market basket analysis.
• There are two important parameters.
• Support
• confidence

6
Method (3/3)

• Cluster analysis
• Each object is very similar to others in the
  cluster.
• Use Hierarchical cluster analysis
• Dendrogram or tree graph.
• EX

7
Data collection

• University professors are important online
  shoppers.
• Gain randomly a list of scholars whose research
  interests are in information technology from
  National Science Council.
• Use search engines to search and count the
  numbers of web pages related to both scholars and
  expertise.

8
Data process

• Filter abnormal data
• Delete the names resulting in extremely large or
  small web page.
• Normalize data
• X scholars expertise
• Y scholars name
• Generate binary data
• If the normalized web page is greater than the
  threshold then
• the binary web page index of X and Y 1
• Else
• the binary web page index of X and Y 0

510
9
Data collection and process result table

• Find out 200 scholars

Y
X
10
Data collection and process result table
1the scholar has the expertise 0the scholar
does not have the expertise

• Find out 200 scholars.

1 0 1 1 0 1 0 1 0 0 1 0 1 1 0 0 0 0 1 .. 0
11
Methodology
12
Model 1 clusters of expertise generated from
association analysis (1/2)

• Each scholar as a transaction and the expertise
  as the item(s)
• The result of Model 1

13
Model 1 clusters of expertise generated from
association analysis (2/2)

• The association map of expertise (shows only 2
  out of 14 clusters)

14
Model 2 clusters of scholars generated from
association analysis

• Each expertise as a transaction and the scholars
  as the item(s)

15
Model 3 cluster of expertise generated from
hierarchical cluster analysis

• Cluster on expertise
• Cluster was set as 30

16
Model 4 cluster of scholars generated from
hierarchical cluster analysis

• Cluster on scholars
• Cluster was set as 30

17
Comparing the differences

• Association analysis (Model 1 2)
• Not every item is appeared on the association map
• Its direct and easier to examine the
  correlations among items on association map.
• Hierarchical cluster analysis (Model 3 4)
• Every item is appeared on the dendrogram
• Its indirect and more difficult to examine the
  correlations among items from the dendrogram.
• The clusters of scholars in Model 4 could aid us
  in designing specific booklists.
• Every scholar belongs to one branch of the
  dendrogram.

18
Result (1/2)

• Evaluate the accuracy of prediction through
  Questionnaire survey.
• Email 2 questions to 200 scholars.
• 42 scholars answered the questions.
• Binomial distribution can be employed to evaluate
  the significance of accuracy rate.

19
Result (2/2)

• Q1.
• Select one most interested booklist from the
  five booklists.
• 1 booklist was the targeted cluster.
• The 4 booklists were randomly chosen from the
  remaining 29 booklists.
• Probability of success is 1/5 20
• Q2.
• Select 5 clusters which were most highly
  correlated with his/her expertise from the 30
  clusters.
• Probability of success is 5/30 16.7

20
Conclusion

• When customers transactional and demographic
  databases are not available, using web mining to
  search potential customers is a good way to avoid
  the limitations of traditional database marketing.

21
Q A

• Thanks for your attention!