Collaborative filtering is a method of
calculating the similarities among a group of customers. The collaborative
filtering application considers site usage behaviors of the group of customers
that have the most in common with the current user as a base for generating
predictions and recommendations for that user. As a simple example, if a
particular customer looks at a book, the collaborative filtering application
determines the group of customers that have either looked at, or purchased the
same book. The application then generates a list of possible recommendations
from the other items that the members of the established group have also looked
at or purchased. The relevance of any particular recommendation is determined
by the number of group members that have looked at the item being recommended.
That is, if 75% of the group have purchased the item being considered for
recommendation, then it would be highly relevant, compared to an item that only
one member of the group has looked at. Finally, the most relevant
recommendations are displayed to the current customer.
Practical implementations of collaborative filtering are more complex than the above illustration. Many use multiple data sources which may include the following:
- Customer-supplied ratings
- Clickstream events that capture the details of a customer's session
- Data from existing company databases
Collaborative filtering is implemented in WebSphere Commerce by integrating Macromedia's LikeMinds software.