Publications
From: Alexander Tuzhilin atuzhili@stern.nyu.edu
Date: Tue, 19 Dec 2000 22:36:28 -0500
Subject: Report on the KDD2000 Panel on Personalization and Data Mining
Report on the KDD2000 Panel
Personalization and Data Mining: Exploring the Synergies
There has been a lot of interest developed recently in personalization
both in the industry and academia as witnessed by several
Personalization Summits (the latest being held in San Francisco in
November 2000) and by the publication of a recent Special Issue of
CACM on personalization (published in August 2000). Personalization
companies focus on building customer profiles, providing
recommendations to the customers and delivering personalized content
and other services. However, there is still no consensus reached in
the industry and academia what personalization really means. To
address this issue, the Panel Chair asked several personalization
experts to provide their views on what personalization means to them,
discuss possible connections between personalization and data mining,
and examine how data mining can contribute to personalization and vice
versa.
The panel consisted of Paul Hagen from Forrester Research, Ronny
Kohavi from Blue Martini, John Riedl from University of Minnesota and
Net Perceptions, and David Zimmerman from YOUpowered. The slides of
their presentations and other pertinent information about the panel
can be found at
http://www.acm.org/sigs/sigkdd/kdd2000/Panel-Personalization.htm
In his presentation, Paul Hagen identified three main functions of
personalization applications: provision of personalized services to
the customer, ability to sell him or her products in a personalized
way, and the support for 1-to-1 marketing. He then argued that the key
for successful personalization applications is the ability to build
rich profiles of customers and ability to deliver personalized content
through the use of content management systems and controlled
vocabularies. He also maintained that today�s personalization
applications are still immature in the sense that they support "point"
applications and focus only on a single marketing channel, mainly the
Web. In contrast to this, 40.9% of on-line buyers use 2 or more and
14.5% use 3 or more channels. Therefore, forward-looking
personalization applications should be able to
- utilize multiple marketing channels, such as the Web, catalogs, and "brick-and-mortar" stores;
- deal with multiple constituencies, such as end-customers, business partners, and employees;
- address different functions, such as marketing, sales, and service.
Paul also argued that personalization applications should be "smart"
in the sense that they should support rich profiles of customers and
provide proactive services to them.
Ronny Kohavi opened his presentation by stating that personalization
is a continuous process of collecting data about the user, creating a
data warehouse for analyzing this data, mining the data from this
warehouse, building customer segments based on the mining results, and
taking appropriate actions targeted towards these segments. He also
maintained that personalization applications should modify the
interaction process with the customer and should not be limited to the
Web but should spread across multiple �touchpoints,� such as
call centers, brick-and-mortar stores, and e-mail. Ronny also
identified different goals of personalization that include
- making customer interactions easier;
- increasing sales by better targeting customer needs;
- saving customer time (he gave an interesting example from Seybold�s
customer.com book of how Motorola customers come to the site, buy products
and exit without wasting any time);
- increasing loyalty and reducing churn.
He also provided various examples of personalization services ranging
from greeting customers by name, to remembering their last shopping
basket, shipping address and credit cards, to changing home page image
and the links. Finally, he presented his personal challenge to the
personalization and data mining communities: personalization
applications should be developed in such as way, that the business
users should be able to understand the analysis results and define
appropriate personalization actions on their own without any help from
the experts.
John Riedl started his presentation with providing interesting
statistics based on real- world personalization applications and then
focused on the future recommendation applications. He maintained that,
in contrast to yesterday�s personalization applications, consumers
will really be in control in the future applications. Marketers will
no longer be able to talk consumers into the deals, control marketing
campaigns, and manipulate consumers in various other ways. He gave
various examples supporting this point. Finally, he presented the
following personalization challenges:
- scalability and real-time performance
- incorporating rich data
- support for consumer-centered recommendations
- development of methods that connect recommenders to marketers.
David Zimmerman started his presentation by defining personalization as the science of
using personal information to uniquely tailor products, content and services to an
individual. Moreover, he categorized personal information into
- PII: personally identifying, such as name, e-mail, and mailing address;
- Non-PII: information that does not identify the individual, such as behavioral
data, imputed data, etc.
He maintained that successful personalization applications require a high degree of PII,
hence an inherent conflict between personalization and privacy. Furthermore, David
outlined major requirements of permission-based data mining systems maintaining that
they must:
- Be designed from the ground up to utilize enterprise-wide privacy models and
protocols;
- Be opt-in;
- Provide mechanisms for informed customer consent;
- Provide several opt-out options, such as a) no data mining of consumer data; b)
data mining for internal use only; c) data mining for both internal and external
uses.
These presentations were followed by several questions from the
audience. As a result, several themes emerged from the panel, such as:
- personalization is a process of collecting and using personal information to
uniquely tailor products, content and services to an individual;
- successful personalization applications should utilize multiple marketing
channels, deal with multiple constituencies and with different functions, such as
marketing, sales and service.
- personalization applications have different goals, such as
- making customer interactions easier;
- increasing sales by better targeting customer needs;
- saving customer time (he gave a very nice example from Seybold�s
customer.com book of hoe Motorola customers come to the cite, buy
products and exit � no time wasted);
- increasing loyalty and reducing churn.
- in contrast to yesterday�s personalization applications, where
marketers would talk consumers into buying products and manipulate
them in various other ways, personalization applications of tomorrow
will empower the customers by putting them in control of the buying
process. Marketers will no longer be able to talk consumers into the
deals, control marketing campaigns, and manipulate consumers in
various other ways.
| |
|