An Analysis of Critique Diversity in Case-Based Recommendation

Critiquing is a well-known form of user feedback in casebased recommender systems. A critique encodes the users preference in relation to a particular feature. For example, in a digital camera recommender a user may be allowed to indicate whether they are interested in cameras with a lower resolution than the one currently presented; so ‘lower resolution’ is an example of a critique over the resolution feature. Recent research demonstrates how the dynamic generation of compound critiques – critiques that operate over multiple features – can deliver significant performance improvements. However user-studies highlight diversity problems that arise during critique generation; for example, one compound critique might constrain resolution, memory and zoom while another might constrain resolution, memory and price. In this paper we describe how critique diversity can be improved and demonstrate that this can lead to significant usability and per-