Does WEB Log Data Reveal Consumer Behavior?

1
Does WEB Log Data Reveal Consumer Behavior?
Faculty of Commerce, Kansai University
Daigo Naito , Kohei Yamamoto , Katsutoshi Yada
, Naohiro Matsumura , Kosuke Ohno and Hiroshi
Tamura
2
The purpose of this research

• Combining various data mining technology ,and
  discovering the new knowledge from Web log data
• from the knowledge that has been gained, planning
  useful sales strategies for future use

3
Background

• Competition between the various shops doing
  business on the Internet is becoming more severe
• -It is necessary to plan the effective sales
  strategies
• Each customers have different purposes and
  actions
• -planning strategies for every different
  purposes and action
• The Web log data that has been accumulated on the
  servers
• -Data mining

4
Explanation of the data
5
Detail of Web log data

• Definition

• Cart It means kosik, and it is also defined
  as purchase.
• Session a session is used as a unit of study
  of a customer.
• PATH it is a procedure of following a route of
  a click made within each site during a given
  session.

6
The data that was subjected to analysis

• Remove the PATH data including only a limited
  numbers and very large numbers of clicks.

A ratio of number of clicks per a session
-A single and 2-4 clicks data does not constitute
enough information for analysis of consumer
behavior. -Session data that included 100 or
more clicks comprised less than 1 of total
sessions and thus, it can be surmised that their
overall importance is not greater.
clicks

• The session data included over 5 clicks and under
  100 clicks (4,220 visitors who made purchase and
  140,327 persons who did not) will be used for the
  analysis.

7
Basic analysis
8
PATH to the purchase
of clicks to the purchase

• PATH to the purchase
• -The number of customers who reached a purchase
  in 7 clicks is the largest.
• In addition, such a customers visit a site and
  purchase it during 2-6 minutes.

of customers
of clicks
Distribution of length of time spent to the
purchase
of customers
Staying time for session (sec)
9
The customer behavior at every each shop site
Differences in Average clicks per session by each
shop

• Differences of
• customers action

-The upper figure shows that differences of
average clicks between every each shop. -The
lower figure indicates that there are the
customers who buy some product categories with a
low number of clicks. But the purchasing visitors
of other product categories use a high number of
clicks. -It is depending on the shop and product
category, customer behavior tends to vary.
of clicks
Differences in number of clicks by product
category
of clicks
10
Strategy for the each shops
11
strategic suggestion for the Characteristics of
each shops

• It would be divided into 3 groups by purchasing
  possibility

Positioning of the shop sites
shop5(MP3)

• Because purchase probability is high, It is
  surmised that the visitors of shop5 have already
  decided what they intend to purchase before they
  visit the shop.
• The strategy that the shop use banners
  advertising to actively induce visitors to come
  to the site can be effective.

Purchasing possibility
of customers
shop3(TV),shop6(mobile)

• Because purchase probability is low, it is
  thought that the visitors of these shops make
  their purchase at regular shops.
• The strategy that involves a joint effort of the
  Click and Mortar strategy type (real shop and
  Net mall shop cooperation) can be recommended.

12
Defining the target shop
Positioning of the shop sites
shop1247

• The purchasing possibility level is about in the
  middle.
• It is possible to raise sales of shop by giving
  purchasing possibility.
• It is necessary to analysis of customer level.

• The number of Shop4s sessions is large.
• We will focus on Shop 4.

13
Strategy for the each customer groups
14
Defining the target customers
We chose to focus on the buying motives of the
visitors to the site. ? the customers that had
already decided on the product they wanted. -The
changing the content of the site would not be
very effective. ? the customers that were
wondering which product to buy. - By putting more
effort into access of the site, it can be
anticipated that we can influence some of them
to purchase.
The difference in the staying time to purchase
of the two groups of customer
We defined customers who went to the cart after
12 strokes or more as customers who were wavering
concerning purchase and designated them as a
segment that required analysis.
Group?
Group?
min
of click
From a figure, the customers who go to the cart
after 12 clicks or more take longer time to
purchase than the customers that go to the cart
in 11 clicks or less.
15
Extraction the rule of target customers
16
Defining the analysis of the objective variables

• the customers that were wondering which product
  to buy. ( the customers who
  go to the cart after 12 clicks or more )
• we extracted the characteristics from among the
  customers who were wavering concerning whether
  to purchase a product or not
• -Target Data
• The visitors of Shop4, among the visitors to the
  site that used 12 clicks or more and also read
  the page of refrigerators-freezers and also read
  the page of refrigerators-freezers
• -the analysis of the objective variables
• -purchase a product(166sessions) or did
  not(346sessions)

E-BONSAI
We extract a rule every customer group.
17
E-BONSAI
E-BONSAI was originally developed to analyze DNA
code. Since then, E-BONSAI has been improved and
by expressing consumer behavior patterns as
character strings, it can be used for extracting
patterns from time-series category patterns as a
data mining tool.
DNA
T
CANCER
A
C
A GAGGCACAGA B GAGTGACAGA C GAGTGACAGA
G
18
the click PATH data convert into character strings
A flow of time
ct
ls
dt
/
ct
popup
Customer A
Mapping Table
Each page is made character string. As follow as
Mapping Table
2 4 5 4 5 1
Customer A

• Characters from the internal site pages can be
  converted into different
• characters and the click PATH data (the data they
  referred to) for all
• visitors that were part of the project can be
  converted into character
• strings.

19
The result of E-BONSAI
177
Searching by functional specification as popup
and findp were used
Mapping Table
Yes
No
purchase!
1555 57
ls (product list) or dt(product explanation) were
seen
(hit/sup)(300/400)
Yes
purchase!
(hit/sup)(28/42)
There are the characteristic such as 2 mentioned
above was seen throughout.
?The factor that we paid special attention was
the multiple searches they made by keying in
terms concerning the functions specifications.
20
Implications for Business
Why do they repeat searching by keying in terms
concerning the functions specifications?
There are two possible reasons for this behavior.

• The page design was bad and it is difficult to
  use the searching function

?There may be a need to improve the design of the
search function page.

• The visitor cant decide that which product
  matches to them

?They need the choice standard for purchase.
Because they dont know what they really want to
purchase.
21
Implications for Business
We suggest that to add a word of mouth reporting
function to a site
With word of mouth information
Evaluation from the user who really bought the
product
22
A Japanese word of mouth bulletin board site
http//www.kakaku.com/
1
A figure of Point count of word of mouth
information
A text search of word of mouth information
2
3
A word of mouth bulletin board
23
Count and comparison of word of mouth information

• Word of mouth-Product comparison by count
  information-Easy to understand!!

The graph of product evaluation by the existing
user
24
Implications for Business
Point count of word of mouth information
A text search of word of mouth information
An at a loss customer
Decision-making support by word of mouth
information
X
X
purchase!