Data mining sales data for Kansei Engineering
In Kansei Engineering (KE), customers are asked for their emotional response to a range of products. KE typically produces 3 dimensional data with customers, products and emotional response via semantic scales as the dimensions. Analysing the customer responses can yield insight into the importance of design factors and the relationships between emotional response and design factors. These relationships are the key to the importance of KE in the design process and in providing a broad portfolio of products. The range of products to be assessed in the KE is best selected using a designed experiment of design factors. The design factors are derived from various sources, including designers and merchandising staff. Data mining and segmentation provide another way of indicating which design factors are important to which types of customers. KE is expensive to do properly, so it is particularly important to prepare the groundwork carefully. This paper investigates what information can be obtained from data mining sales data as a pre-cursor to Kansei Engineering.
| Attachment | Size |
|---|---|
| Coleman IPROMS-2 2007.pdf | 147.38 KB |
| S.Coleman.pdf | 147.04 KB |
Dear Prof. Pham,
Thank u for ur question, here below is the reply i received from Dr Shirley.
"Commercial businesses which provide database segmentation are not always mindful of the behaviour of customers for the company in question (as was our experience). Where this is the case, it can mean that the database is grouped into segments which are too large to be meaningful. For example, external companies may benchmark active customers as those
with order dates in the last 24 months. If 50% of your atabase is outside of this range it would be meaningless to group so many customers into one segment. Therefore being aware of the range and distribution of data is essential for getting the most out of the database.
Once a scoring system for segmentation has been decided on, it is possible to run a mock scenario of 'what would have happened' by segmenting customer data which excludes purchases in the last 6 months. The subsequent purchases in the most recent 6 months can then be matched back to the segmentation to ensure it yields the expected results. From
our experience this level of investigation is not routinely carried out by external companies. "
Dear author:
Thank you for the paper. As you mentioned the segmentation is based on statistical distribution. Do you use any parametric model distribution model such as t, gaussian or gamma model or using non-parametric distribution to determine the density in customers for segmentation?
Hi Dr Charles!
Many thanks.
Have a very good trip home.
We look forward to seeing you again soon - if not in person, then at least virtually at IPROMS 2008!
Best wishes.
DTP.
Thank you, professor.
Reached home safely, Will see u soon, at least virutally as u said.
best regards,
Charles.
Dear Charles,
It's good to hear from you and to know you have reached home safely.
See you again soon!
Best wishes.
DTP.
PS: We hope you have told everyone about the Cardiff Bay Bees!
Dear authors,
Can you please elaborate on the following statement?
"Basing segmentation on statistical distributions is a novel approach; it is unusual to be able to validate segmentation using past data and current sales."
Why is it unusual to be able to validate segmentation using past data and current sales?
I would have thought it quite normal to use past data to perform segmentation and then validate the results using either a different set of past data or current data or a combination of past and current data?
Thank you.
D Pham.