product recommendation and feedback mining
TRANSCRIPT
Mahak Gupta(10103496)
Mentor – Ms. Adwitiya Sinha
INTRODUCTION
Online shopping has emerged as the newest big thing andwhy not?It’s easy. It’s safe, and the best of all it saves TIME!Providing personalized product recommendations forshoppers on ecommerce sites has been proven to boostorder values, increase customer loyalty and enhance theonline shopping experience.
OBJECTIVE
The objective is to develop an application that will providethe online shopping customer a specific range of productscustomized according to their previous preferences andcharacteristic constraints through a website with highdatabase handling capability and also taking into accounttheir suggestions and reviews and giving them the assurancethat their opinion also matters.
LITERATURE SURVEY
We researched several papers and studied them thoroughly to
understand our topic and decide the path to implement it.
There were main 8 papers we selected to draw our work from.
They were in 2 broad categories namely,
online shopping with its applications and processes
Data mining algorithms to analyze the data and extract
results
Online Shopping Portal Processes
Data Mining Algorithm Operations
OPEN PROBLEMS AND ISSUES
Fickle-mindedness of the customer
Customer input
Real time nature of searches
Accuracy of the characteristic data
Multiple entries of same product varying on some
characteristic
Segmenting feedback based on phrases
NOVELTY AND BENEFITS
It acts as a personal shopper. It takes into account both previous
searches and current preferences for current real time
recommendations simultaneously.
Since it gives you the user ratings rather than the reviewer or
company ratings so that a realistic idea can be determined not the
hyped up image set by the manufacturers.
Segregation of feedback text is done on the basis of the complete
meaning of the phrase and not just individual words. Eg. ”Not Bad”
It is important to take into account the customer feedback as the
product is only worth what the customer sees it as.
PROPOSED ALGORITHM
As our program is about finding the right product to recommendto the customer based on their previous searches and preferences,we realized that one particular algorithm or method would not bethe right approach for us. So we decided to take some of thepopular algorithms of data mining and streamline them accordingto our requirements.
Naïve-bayes algorithmApriori algorithmK-means algorithmSoundex AlgorithmEtc.
TOOLS AND TECHNOLOGY
Microsoft Visual Studio 2012 Ultimate
C# with .NET Framework
MS Access for database
Characteristic Jabong Flipkart Myntra Snapdeal US
View without
registration
yes Yes yes no Yes
Shopping Cart Yes Yes Yes yes No
Recently
viewed
Yes No No No Yes
Constraint as
per categories
yes Yes no No Yes
Comparison of Other Existing Approaches/
Solution to the Problem Framed
IMPLEMENTATION
LAYOUTS
Forms for user interactions were made in visual studio using c# for
login, registration, password change, product display and selection, and
feedback gathering.
DATABASES
Synthetic database for products with their characteristics and each
user’s previous visits as well as their choices was created.
Also ratings and quality of manufacture is also added in the tables per
product. Table is also created for user password and login and also for
customer details.
IMPLEMENTATION
EXTRACTION CODE
For feedback mining real feedback data is extracted from xml file into
MS Access database for further analysis using an extraction code in C#
ALGORITHMS
Naïve-Bayes algorithm and K-means are data mining algorithms used
for classification and clustering of data. We have implemented it on our
synthetic database of online products to classify them as
recommendable or not as per each users preferences. We have also
implemented Soundex for feedback mining.
TEST PLAN
The purpose of testing is quality assurance, verification and
validation, or reliability estimation.
Unit Testing
Component testing
Integration testing
Validation Testing
System Testing