Personalization Techniques - Page 2

There are a few generally known techniques for personalization, and we'll discuss the pros and cons of each here.

The first is rules based personalization - which is probably the easiest to understand and implement. Designers must know ahead of time what the condition is, what to do about it, and it is often similar to an if/then type format. For instance, Business A knows that they have printer paper overstocked, so they decide they need to get rid of it somehow. So, if a customer adds a printer to their "shopping cart," then prompt them if they want to buy some printer paper. The business could also incorporate sales or discounts in this approach - if a customer buys a printer, we'll sell them paper at half price.

The drawbacks to rules based personalization are its limited applicability. The key to this scheme is that the developer must know ahead of time what the promotion or personalization should be. This is rather strict, and must constantly be evaluated and adjusted depending on the business's needs. The scale of this scheme can also be very large if more detailed personalization is required.

A second method of personalization is collaborative filtering or community based. This method offers users customized information depending on what others like him have done previously. This is the type of personalization you see when Amazon.com tells you "People who bought this book also bought..." Based on what previous users have done, we tailor the site for you. This is a very general way to personalize that may often present incorrect customizations, but more often it is of some help to the user.

The problems here are that this scheme must make broad generalizations, and that it cannot leverage any personal information, as only large demographic data is evaluated. Also, this scheme relies on having information from previous users. If no other users have given you any data to use, then future users won't receive anything meaningful, if anything at all.

Finally, a third method that is just starting to be used in the industry is inference. This method relies on learning the user's behaviour by monitoring what they're doing at the moment. For instance, we monitor a user's clicks and see that they've clicked on computer books the last 7 out of 10 times. Therefore, we infer that this user is interested in computer books, and we can tailor the site accordingly. Examples would be to prompt them if they'd like a discount on computer book, or for more information, or modify the home page so that it displays featured computer books. This method is similar to rules based, with one key difference that we'll talk about in a moment.

Another "feature" of this model is its ability to learn a user's behaviour. Seeing that they click on computer books may mean that they are a computer book shopper, and can be treated as such by the application. If, on the other hand, they don't click on any pattern of items, we can treat them as a window shopper, and adjust accordingly. By lumping users into categories, developers can start to predict their needs.

This scheme, obviously, requires the most resources of the three techniques. While not extremely difficult, it does consume resources to monitor the clicks for every user that comes along. When incorporating personal information (i.e. only show the computer book discount if they've bought a computer in the past 90 days), then this scheme demands even more resources. Another drawback is that developers must also need to know the rules ahead of time, though they need not be as complete as with rules based. While this method could deliver the highest degree of personalization, it may be too cumbersome to use all of the time.

We can see that there are advantages and drawbacks to each method, but the first two are characterized by non-learning systems and non-real-time user feedback. Real time feedback would be in the form of an immediate reaction of the application to a user's interaction, rather than a prescripted behaviour. The third method can overcome these drawbacks but relies on using a large amount of data and resources.

So with these three schemes, we cover pre-meditated personalization, demographic personalization, and spontaneous personalization - all of the important stages of personalization. Now let's take a look at the implementation of these schemes.

Everything You Need to Know About Personalization - Page 1
Everything You Need to Know About Personalization - Page 1
Personalization Implementation - Page 3