Measuring similarity of users with qualitative preferences for service selection

Similarity measures are essential in many preference-based personalized applications such as collaborative recommendation and web service selection. However, previous studies have been mainly focused on the similarity measures for quantitative preference rather than those for qualitative preference, though the latter has attracted much attention recently. This paper aims to fill in this gap by proposing an intuitive similarity measure for conditional qualitative preference which is represented by CP-nets. In particular, we introduce two methods, a basic and a general similarity measures, corresponding to whether two CP-nets share similar structures and contents or not. Experimental results on two real-world data sets demonstrate that our similarity measure can not only correctly reflect the changes of users’ preferences, but also be effective in identifying similar users. In addition, only by adopting the K most important attributes, the computational cost can be greatly reduced while sufficiently high accuracy is preserved. Furthermore, we demonstrate the effectiveness of our method in complementing users’ preferences by aggregating those of similar users in a scenario where users’ preferences are incomplete.