Transcript Does WEB Log Data Reveal Consumer Behavior?

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
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
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
Explanation
of
the data
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.
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 47%
clicks comprised less than 1% of total
sessions and thus, it can be surmised that
their overall importance is not greater.
3%
49%
1
2-4
5-99
100-300
clicks
1%
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.
Basic analysis
PATH to the purchase
400
# of clicks to the purchase
350
# of customers
300
PATH to the purchase
250
200
‐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.
150
100
50
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
# of clicks
300
200
Distribution of length of time
spent to the purchase
150
100
50
0
0-59
60-119
120-179
180-239
240-299
300-359
360-419
420-479
480-539
540-599
600-659
660-719
720-779
780-839
840-899
900-959
960-1019
1020-1079
1080-1139
1140-1199
1200-1259
1260-1319
1320-1379
1380-1439
1440-1499
1500-1559
1560-1619
1620-1679
1680-1739
1740-1800
# of customers
250
Staying time for session (sec)
The customer behavior at every each shop site
Differences in Average clicks per session by each shop
35
30
# of clicks
25
Differences of
customers action
‐The upper figure shows that
differences of average clicks
between every each shop.
20
15
10
5
0
shop 1
shop 2
shop 3
shop 4
shop 5
shop 6
shop 7
Differences in number of clicks by product category
100%
90%
80%
‐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.
70%
60%
# of clicks
50%
25～
20-24
15-19
10-14
5-9
0-4
40%
30%
20%
10%
0%
D igitalcam eras
Film cam eras
W ashingm achines
D ishR efrigerator C ookersw ashers
s,and-ovens
freezers,show -cases
‐It is depending on the shop
and product category, customer
behavior tends to vary.
Strategy for the each
shops
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）
Purchasing possibility
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.
# of customers
Shop1 camera
Shop2 audio
Shop3 TV
Shop4 electrical appliances
Shop5 MP3
Shop6 mobile
Shop7 PC
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.
Defining the target shop
Positioning of the shop sites
shop1・2・4・7
Purchasing possibility
The purchasing possibility level is
about in the middle.
It is possible to raise sales of shop by
giving purchasing possibility.
# of customers
Shop1 camera
Shop2 audio
Shop3 TV
Shop4 electrical appliances
Shop5 MP3
Shop6 mobile
Shop7 PC
It is necessary to analysis of customer
level.
The number of Shop4’s sessions is large.
We will focus on Shop 4.
Extraction the rule
of target customers
Defining the analysis of the objective variables

the customers that were wondering which product to buy.
customers who go to the cart after 12 clicks or more
（
the
）
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 refrigeratorsfreezers
‐the analysis of the objective variables
‐purchase a product（166sessions） or did not（346sessions）
E-BONSAI
Ｗｅ extract a rule every customer group.
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
A
C
G
CANCER
A GAGGCACAGA …
B GAGTGACAGA …
C GAGTGACAGA …
the click PATH data convert into character strings
A flow of time
Customer A
/
ct
ls
ct
dt
popup
Mapping Table
Each page is made character string.
As follow as Mapping Table
Customer A
245451
category alphabet
popup
1
/
2
faq
3
ct
4
ls or dt
5
m ailc
6
findp
7
•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.
The result of E-BONSAI
1*7*7*
Mapping Table
Yes
No
category alphabet
popup
1
purchase!
/
2
1*5*5*5
faq
3
*5*7*
ct
4
(hit/sup)=(300/400)
ls or dt
5
Yes
m ailc
6
findp
7
purchase!
Searching by functional
specification as popup and
findp were used
ls (product list) or
dt(product explanation)
were seen
(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.
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 can’t decide that which product matches to them
⇒They need the choice standard for purchase.
Because they don’t know what they really want to purchase.
Implications for Business
We suggest that to add a word of mouth
reporting function to a site
With word of mouth information
Information from a maker
An instruction from a site
Evaluation from the user
who really bought the product
A Japanese word of mouth bulletin board site
http://www.kakaku.com/
1 A figure of Point count
of word of mouth informatio
2 A text search of word
of mouth information
3
A word of mouth bulletin
board
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
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
purchase！
Ｘ
Ｘ