The Ins and Outs of Critiquing

Eliminating previously recommended products in critiquing limits the choices available to users when they attempt to navigate back to products they critiqued earlier in the dialogue (e.g., in search of cheaper alternatives). In the worst case, a user may find that the only product she is prepared to accept (e.g., having ruled out cheaper alternatives) has been eliminated. However, an equally serious problem if previous recommendations are not eliminated is that products that satisfy the user’s requirements, if any, may be unreachable by any sequence of critiques. We present a new version of progressive critiquing that leaves open the option of repeating a previous recommendation while also addressing the unreachability problem. Our empirical results show that the approach is most effective when users refrain from over-critiquing attributes whose current values are acceptable.

• Type: Conference
• Year: 2007
• Where: IJCAI
• Authors: David McSherry, David W. Aha
• Keywords: Artificial Intelligence, Cheaper Alternatives, IJCAI 2007, User’s Requirements, Worst Case