Types of Personalization

Description of Personalization filtering, and examples of when to use a filtering type.

There are three types of Personalization:

Simple filtering

A site displays content based on predefined groups of site visitors. For example, if a site visitor is in the Human Resources department, the site provides access to URLs containing Human Resources policy manuals.

Rules engines

In a rules based system, the site owner defines a set of business rules which determine what category of content is shown when a certain profile type visits the site. An example would be: Display all four wheel drive SUVs to visitors in the northeast in the 21 to 35 age group.

This approach has the advantage of driving the site's behavior with the business objectives of the site owner. The site owner is usually the owner of a marketing campaign or some other business manager.

Collaborative filtering

A site visitor rates a selection of products, explicitly or implicitly. Those ratings are compared with the ratings offered by other visitors. Software algorithms detect similarities. For example, a visitor receives book recommendations based on the similar purchases of others.

 

Rules versus collaborative filtering

When complex filtering is required, a rule-based system may work better than collaborative filtering, and vice versa. The following table details examples where one type of personalization is better than the other.

When to use rules filtering versus collaborative filtering.
Scenario Which filtering type to use Reason
If the number of items offered and users who purchase them are rather low. Rules Very little room to compute user similarity necessary for collaborative filtering.
If price points are high or purchasing frequency is low. Rules Finite, limited arenas - collaborative filtering fails because of the inherent lack of diversity.
If there is a pre-existing dependency between items. Example: Disability policy required for homeowner Rules Recommending a disability policy just because collaborative filtering says many others "like this user" also bought a policy is incorrect--one must have the homeowner policy first.
If number of items offered and users who purchase them are rather high. Collaborative Cannot write rules covering all items.
If price points are low, all quite dissimilar, or the products offered have a wide range of user appeal. Collaborative The wide variance fits the collaborative filtering approach. Collaborative filtering also lowers the risk of making "bad" recommendations.
When not much information is gathered about the user, but the user can be identified, possibly by a login or cookie. Collaborative In this case, user attributes on which to base rules may be lacking. Collaborative filtering can compare the user's experiences on the site to other users.

Home |