Collaborative filtering is a relatively new technique to the database marketing field, gaining popularity with the advent of the Internet and the need for “recommendation engines.” We discuss the two major forms of collaborative filtering: memory-based and model-based. The classic memorybased method is “nearest neighbor,” where predictions of a target customer's preferences for a target product are based on customers who appear to have similar tastes to the target customer. A more recently used method is itembased collaborative filtering, which is model-based. In item-based collaborative filtering predictions of a target customer's preferences are based on whether customers who like the same products the target customer likes tend to like the target product. We discuss these and several other methods of collaborative filtering, as well as current issues and extensions.
KeywordsActive User Target User Collaborative Filter Preference Rating Rating Threshold
Unable to display preview. Download preview PDF.