Personalization Simulation - Page 4
November 22, 2000
Before we go on any further, it is important to note again the focus
of this article. Many of our examples have applied to e-commerce
concepts, simply because these settings are the ones to benefit
most from personalization techniques. However, these techniques
can definitely be applied to other types of applications, such as
for a news web site, a web portal, or even a desktop based
application such as Microsoft Word (which, by the way, uses some
of these techniques in version 2000 and above). The first part
of this article is meant to outline concepts and techniques, rather
than specific implementation details, which will be discussed in the
next section.
Simulation
Simulation is the last part of the concepts we will illustrate here,
but it is an important one, especially for the inference scheme.
Simulation does not mean testing all code to ensure it is bug free
or that it does what it intends to do. Rather, simulation is predicting
what a user will do, and how to react to it.
For example, say we develop a rule to push printer paper to any
customer who buys a printer. After our application has been
running for some time, we take a look at the purchase histories, and
we notice something interesting. Though every single user who
bought a printer was presented the offer to buy printer paper
at a discount price, not a single user actually purchased the
paper. We have predicted that printer buyers would also buy
printer paper, and we were wrong. Therefore, this rule has done us
no good, and may have actually deterred some customers from making
purchases (how many times have you tried to buy something online only
to change your mind once you are swamped with "additional offers" or
pop-up windows?).
In a non-e-commerce setting, imagine you want to get users to
register and keep visiting your web portal. You think that if a user
registers, they would obviously be interested in getting email alerts
on their favorite sports scores. However, none of your users ever
signs up for the sport scores alerts, and it may have driven users
away from your site, because they feel pressured, or are just annoyed
with your tactics. Again, simulation here would have helped.
Simulation ensures that the predictions we make are actually correct,
and though it is not a necessary step in personalization, it can
be very helpful, especially for e-commerce settings. Simulations take
information from every possible source - purchase history, demographic
information, clickstream, etc - and try to predict what would happen
in a given situation. It may sound like a very imprecise science
prone to many errors, but many businesses are actually reporting
increased revenue because of these simulation and personalization
techniques.
We base our rules on the results from the simulations, which look
at what a user has clicked on recently, what they've bought previously,
and what special purpose group they fall into. A simulation could
tell us obvious information such as that a user who's bought five
computer books in the past is more likely to buy another one than
someone who has never bought a computer book, to more complex information,
such as that users who click on image links rather than text links,
and spend an average of three minutes or greater per web site are more
than three times as likely to buy clothes online than others. (Note
that this information and numbers are completely hypothetical, and
do not represent any real world information that this author is
aware of).
We're going to skip over many of the implementation details for
simulations, but in essence, the key to successful predictions are
analyzing probabilities based on what you already know about someone.
Measure the amount of times something successful (such as buying a
computer book, or printer paper) has happened, versus the rest of the
user's information, and you should arrive at a good estimate.
In summary, simulation is a very helpful tool that should be
considered in any personalization undertaking.
Personalization Implementation - Page 3
Everything You Need to Know About Personalization - Page 1
A Few More Considerations - Page 5
|
