Browse Prior Art Database

Method for intelligent segment user group to delivery of personalized informational and transactional data by data mining

IP.com Disclosure Number: IPCOM000244451D
Publication Date: 2015-Dec-13
Document File: 5 page(s) / 155K

Publishing Venue

The IP.com Prior Art Database

Abstract

This disclosure is using logistic regression, remove attributes which do not have statistical effect on customers purchase behavior. customer segmentation rules by using decision tree, and statistic the purchasing probability of each customer group are used in the disclosure.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 48% of the total text.

Page 01 of 5

Method for intelligent segment user group to delivery of personalized informational and transactional data by data mining

Nowadays, the work flow for sending marketing emails to target customers is: first, upload customers' data, setup customers' attributes; then, setup customers' groups/rules manually; after that, setup emails sending rules; finally, send out the emails automatically.

It's really convenient for clients to email costumers, but there is further improvement on the automation and user experience.

Problem 1: Setup customers' groups/rules manually.

Complaint from a user: it need create new groups/rules manually. -When the client has new target customers.

Problem 2: The accuracy of the groups/rules.

Complaint from a user: There is lack of scientific basis of the groups/rules

-When the groups/rules are made by the clients' order.

-When the groups/rules are depended on peoples' thoughts.

1. Elimination of irrelevant customer attributes:

By using logistic regression, remove those attributes which do not have statistical effect on customers purchase behaviour.
2. Intelligent customer segmentation:

Create customer segmentation rules by using decision tree, and statistic the purchasing probability of each customer group.
3. Customized threshold:

Send promotion Emails to potential customers who meet customized threshold.

In the old method, its workflow works like below:

In this new method, the intelligent recommendation by data mining is introduced. And now this disclosure make it more smart in this flow..

1



Page 02 of 5

And in this process, there are two steps:

Step

Step
111::: use logistic model to filter those important variables which have high influence for purchasing power

use logistic model to filter those important variables which have high influence for purchasing power. .

Given the history data, if a customer with certain properties purchased a product. Note y is the dependent variable. If the customer purchased, its value is 1, or it's 0.

Supposed the following properties will affect if customer purchases that product again. And then use this logistic regression model to estimate coefficients

2

Also T test will be used to test if these relevant variables are significant. For those high important properties, they will be used in the Decision Tree Model.



Page 03 of 5

Step

Step

222::: Get the customer segmentation rules and corresponding purchasing probability by using decision tree Get the customer segmentation rules and corresponding purchasing probability by using decision tree .

For example:

There is database as follows: messages of the customers(Age, Work,etc) and the last column is the purchase, if the purchase is "Y", it

means this customer bought this product

, and "N" means not bought.

Ag e

education-nu m

native-countr y

Incom e

Purchas e

work

education

marital-status

occupation

relationship

race

sex

 Not-in-famil y

Y

39

 State-gov

 Bachelors

13

 Never-married

 Adm-clerical

 White

 Male

 United-States

 ...