Paper for MIS531A Team Project

overratedbeltAI and Robotics

Nov 25, 2013 (3 years and 9 months ago)

90 views


1


Paper for MIS531A Team Project

I
nstructor:
Dr
.

Hsinchun Chen




PriceSmart.com



Fall 2005

Kaijia Bao


Manlu Liu

Sandi Wenas


Tianjun Fu


2


Index


Introduction

................................
................................
................

3

Business Model

................................
................................
...........

4

Functionality

................................
................................
...............

5

Customer analysis

................................
................................
.................

5

Market analysis

................................
................................
.....................

7

Product analysis

................................
................................
....................

8

Revenue analysis

................................
................................
.................

10

Development Architec
ture

................................
......................

12

Application Architecture

................................
.........................

12

Amazon API

................................
................................
........................

13

Database Desig
n

................................
................................
.................

14

Data Mining

................................
................................
........................

15

Visualization

................................
................................
.......................

17

Interface

................................
................................
..............................

17

Novelty

................................
................................
.......................

18

Assignment

................................
................................
................

18



3

Introduction

PriceSmart
.com

is a web business
tool
which conducts web
-
based analysi
s base
d

on
information from Amazon.
Currently, there is no tool available in the market that
provides unique value as PriceSmart.com offers. Most of the tools that are available
such as wishlist, suggested items to buy, top 5 items, price comparison of ite
ms on
different vendors, only provide convenience for buyers.


PriceSmart.com

was designed to
focus

mainly
on sellers instead of buyers.
We believe
that it is important for the sellers to understand their market i.e. what does the
customer like, what char
acteristics affect the sales of an item the most, etc.
These are
critical
information

for sellers to
increase

their
revenue
.
PriceSmar.com
provides
thorough analysis on market situation for small businesses that make profit from
selling their items through

Amazon.


Amazon
is one of the most popular

companies

in the world.

Amazon experienced
several milestones, which indicated the rapid
development

of Amazon business.
Amazon started its
business

in 1995; in 1998, Amazon
launched

new
categories

and
entered E
urope market; in 1999, Amazon launched auction and commerce network; in
2000, living.com declared bankruptcy, Amazon built relationship with toysrus.com
.


As we can see, Amazon successful
ly developed its business model
.

T
he number of
Amazon
website
visitor
s
is
growing rapidly
.
Thus
,
t
he historic data
that provides
over

4

three years of past sales data
become more and more
valuable. Currently Amazon
charges $400 per month for historic pricing

only for accessing the data
. It will be a
burden for small
businesse
s using Amazon historic data.

Our idea was generated base
on this niche market.


Business Model

PriceSmart.com

not only
shows
historical pricing from Amazon, but also conducts
enrich analysis.
As
we mentioned

above, o
ur potential customers are small busine
sses
who
are interested in
making money from selling their items from Amazon. They will
be able to get a thorough analysis from their past sales to help them in making
informed decision to increase their revenue.


T
h
e potential benefits can be considered i
n the following ways:

1.

PriceSmart.com

can build partnership with Amazon.
PriceSmart.com

uses the data
retrieved
from Amazon, and conducts
thorough
analysis based on these data.
Amazon might

be

interested in
providing
this service
for
its customers

as an
add
ition to historical pricing data that they provide
, which
then will
lead to
partnership with
PriceSmart.com
.

2.

PriceSmart.com

can charge monthly fee or membership fee from our clients.
C
omparing with fee Amazon charges to its clients, our fee would be very
competitive to attract small businesses.

3.

If PriceSmart runs very well, we will look for benefit from advertising.


5

The above three benefits can only be
received

when
PriceSmart.com

is well
recognized. So in the first stage, we will put main effort to find v
enture capital to
support. Eventually we target at moving our business to IPO.


Currently
, due to time limitation,
we
only
focus our
analysis
in books. We expect to
expand
our business model
to include other categories including electronics,
CDs,
camera, c
omputer
,

etc
.
There are
already
lots of web businesses regarding book

available in the market
. Most of them focus on buyers.
T
he Unique value we provided
to our clients is enriched analysis of Amazon data, which mainly focused on sellers.


Functionality

Th
ere are mainly four types of analysis in our business functionality.

They are:



C
ustomer analysis



M
arket analysis



P
roduct analysis



R
evenue analysis


Customer analysis

In customer analysis,
the clients

can understand who
their

customers
are
and
their
purchas
ing

behavior.

We
rank
the
top five customers
of
our
clients’
current customers.
The ranking is based on the purchasing times. From this analysis, we can
also
define
our clients’
potential customers.

Furthermore, we

also rank 5
top
desired genres. This

6

info
rmation will be useful for sellers who want to
increase

sales.

Moreover, w
e
also
provide customer list which includes summary information, genre interests and wish
list items for every customer.
T
here is links
available
to
direct the customer to
Amazon for

a particular wish list item
.

(See Figure1, Figure 2)


Figure 1 Buyer Analysis


Figure 2 B
u
yer Details


7

Market analysis

Market analysis will provide the whole picture of this market.
It will help
our clients
to understand what

is happening
on
the
market,
and pursue potential market
opportunities.


We analyze the total market size, list the
5
top desired items
and

list the
5
top desired
genres.

This information received from market analysis can help
our clients
to define
the opportunity for expanding
their
current market into
new product

lines

in the future.

W
e also group the
information

generated from wish list into different clusters. The
distance
between

clusters indicates

different

level of relationship.

(See

Figure 3

and
4
)


Figure 3 Cluster Analysis


8


Figure 4 Market Analysis


Product analysis

In product analysis, we define items status as three types: open items, closed items
and canceled i
tems
. In this example,

from the ID3 analysis provided by Analysis
Service and Business Intelligence Studio 2005,
we found out that the
List price, start
date, Amazon FK and sub condition are four
strongest
factors
that determine those
condition
.

If
our clients still want

to know more detailed information, we provide
the Neural Network analysis to
help them to
unders
tand which combination of
attributes
is
going to
generate the best result, f
or example, selling books that are in
good condition right before the thanksgiving break.


Finally, product analysis also provides our clients with prediction of the current open
item. Given with all the attributes that they currently has, which of item are more
likely to be closed (sold) and which of them are likely to be canceled.
With

these

9

complete
understanding of the market
and thorough analyses,
our clients
can gain
competit
ive advantage
. This in turn, will help them to generate higher revenue. (see
figure 5, 6 and 7)




F
i
gure 5 Product Analysis


10


Figure 6 Neural Network


Figure 7 Items Status Forecast


Revenue analysis

In revenue analysis, we list the
5
top genres by qua
ntities and the
5
top

genres by
revenues. Historical revenue data is listed and deviations are showed in revenue
forecast.

R
evenue analysis can help
our clients
to
understand
factors
that
affect
their
revenue and find the way to improve
them
.

Statistics an
d information visualization are

11

used in
these
functionalities.
Moreover, t
he graphs
that
we provide are user friendly.

(See F
i
gure
8
)


Figure 8

Revenue Forecast


Business Value

With all of these functionalities, we believe that PriceSmart.com
can provide
great
business

value for our clients.

1.

PriceSmart.com

can help our clients to understand their market segment and gain
competitive advantage.

2.

PriceSmart.com

can help our clients to understand who their current customers
are
and what their customers look lik
e
,

so that they can better serve their customers.

3.

PriceSmart.com

can

also
help our client
s

to understand
their
potential customers
and new market opportunity.

4.

PriceSmart.com

offers
simple and efficient interface for our clients to manage
their customers an
d products easily.


12

5.

PriceSmart.com

can help our clients to
understand characteristics of current
market and find the best combination that affect the sales. That in turn, can help
them to decide when they want to
expand
their

product line into other profit
able
areas
in their market segment
. Altogether as a function can
help
our clients
to
improve
their
sales revenue.

6.

PriceSmart.com

can help our clients to understand
their

revenue model and predict
future revenue.
With this function, our clients will be able

to decide when the best
time for them to sell their items is.


Development Architecture

IDE:



Visual Studio 2005

Web server:

Internet Information Service 6.0

Data Base:


MS SQL Server 2005

Framework:


ASP .NET 2.0

Language:


C#


Application Architecture

We first collect data from Amazon using the API
. Then, after Amazon returned the
data, we
save
them
to our database. After finishing
in
collecting
the
data, we
transform our data into those that can be used by SQL 2005 data analysis algorithms
or directly

pulled into the web page. Then we embed windows user controls, which
show the data mining analysis result, into out website in a similar way as ActiveX


13

does
. In the following part, each component will be introduced in detail.

(See Figure
9)


Figure 9

Arc
hitecture

Amazon API

We use the API of Amazon
E
-
Commerce Service (ECS) to collect our data. ECS
exposes Amazon's product data and e
-
commerce functionality. This allows developers,
web site owners and merchants to leverage the data and functionality that Am
azon
uses to power its own e
-
commerce business.


The data we collect contains information for Amazon customers, sellers, products and
sales history. Details will be
provided
in the following Database session. However,
there are several limitations for usin
g Amazon API. For example, every day we could
only collect information for 50 Amazon users and sales history for all Amazon sellers
was not available until recently Amazon released its “
Amazon Historical Pricing”,
which will cost people more than $400 per
month to use it. Therefore, we have to

14

collect data daily and use the sales history of those sellers whose information we have
collected to do further data mining.


We follow the steps that we have learned in the lab session to collect data. The only
diffe
rence is that we use SOAP instead of REST for retrieval of those data. It is much
more convenient to classify the data we collect by using SOAP. It allows us to
encapsulate the information
collected
by using Object.


Database Design

Here is our database De
sign. We have defined six tables to store our data

(see Figure
10)
.

Table "Amazon_People" contains basic information for Amazon users that we
have collected. Table "Amazon_WishItems" describes items collected in customers’
wishlists. Table "Amazon_Feedback
s" records feedback from and to both buyers and
sellers. Table "Amazon_Products" contains basic information for products on Amazon,
which in our case is the book. Table "Amazon_offers" describes the sales history of
open and closed items on Amazon. Table "
Amazon_BrowseNodes" stores the
category of each product that the buyers bought.


15

Amazon_BrowseNodes *
ProductFK
NodeName
Amazon_Feedbacks *
SellerFK
BuyerFK
FeedbackDate
Amazon_People *
AmazonPK
Name
AverageRating
City
State
Amazon_Products *
ProductPK
Authors
Title
EditorialReview
DetailsURL
AverageRating
ListPrice
Availability
Amazon_WishItems *
AmazonFK
ProductFK
QuantityDesired
QuantityReceived
DateAdded
Amazon_Offers *
AmazonFK
ProductFK
Condition
SubCondition
Price
StartDate
EndDate
Status
CloseDate

F
i
gure 10 data table


We use MS SQL Server 2005 for our database. The reason why we use it is that it
provides powerful data processing and contains powerful data mining algo
rithms that
we can utilize.

However, there is some limitation with the edition that we used. We
currently use MS SQL Server 2005 Developer Edition. This edition is limiting
anyone
who wants
to access our website
to the local machine
. Furthermore, the reaso
n why
we do not use the enterprise edition was because we do not have the access to the
program and it only runs on Windows Server edition.


Data Mining

We use Microsoft Analysis Service and Business Intelligence Studio 2005 to do the
data mining part

(see

Figure 11)
.

Different algorithms have been used for different

16

purpose:



In market analysis, we use clustering algorithm to cluster customers into
several group, which help sellers identify their potential customers.



In product analysis, we use clustering
algorithm to cluster multiple products,
which can help sellers identify the most profitable as well as most welcomed
products; we also use ID3 algorithms to analyze the function of different
attributes in sales. Therefore, if a seller finds out that seller

rating plays the
most important role in successful sale, he can puts improving his seller rating
into his top priority.



Figure 11 Data Mini
n
g Lift Chart




In product analysis, we also predict the probability that a product will be
"sold" or "canceled".
We have tried ID3, Cluster and Neural network

17

algorithm. Finally, we found that Neural Network algorithm provides us with
the highest accuracy. Therefore, we used neural network as the prediction
algorithm. Above is the accuracy comparison result.



In rev
enue analysis, we use time series algorithm to predict future profit
based on the current sales history of the seller. The seller can further use the
information to predict when the best time for him to sell his items is.


Visualization

Using a unique tech
nique that allowed embedding of Windows control into Internet
Explore, we were able to visualize the information given to us by t
he Microsoft
Analysis Service.
The f
ollowing links is the tutorial:
http://www.devhood.com/tutorials/tutorial_details.aspx?tuto
rial_id=187


Using the process described in the tutorial, we were able to inherit some of the
visualization controls for windows offered to us by Microsoft and add the ability for it
to retrieve the necessary information from the Analysis Service to visu
alize the
appropriate data.


Interface

We use ASP.NET to design our interface. It is simple but user
-
friendly. One main
feature is interactive. That is by embedding the data mining result into our websites,
we can let users select
on
the data mining result

if they want to get further

18

information regarding that specific attribute.


Novelty

Our main novelty is that we provide a comprehensive market analysis using aggregate
statistical analysis and multiple data mining algorithms such as clustering algorithms,

ID3 algorithms, neural network algorithms, time series algorithms, etc. We have also
explored methods to embed windows control on the web and successfully implement
it. Some minor novelty includes that we implement an interactive and user
-
friendly
visuali
zation and we have utilized object oriented design in our implementation.


Assignment

Kaijia Ba
o


Database, Data mining, Webs
ite, Visualization, Presentation

Tianjun Fu


API, Presentation, Paper

Manlu Liu


Team Coordination, Website, Presentation, Paper

Sandi Wenas


Database, Website, API, Presentation, Paper