Application of clustering in modeling user preference for leisure

Multiple commercial systems focus on recommending suitable leisure activities to their users, yet little academic research has explored how to describe and model user preference for leisure. This paper investigates dimension-based and clustering-based approaches to modeling such preference. We rated a set of 135 leisure activities on 17 leisure dimensions in a user study, and relied on these dimensions as feature vectors to cluster the 135 leisure activities using k-means and its fuzzy variant. Another user study collected user preference for the 135 activities to be used as the ground truth. The 17 dimension and resulting clusters were then tested as descriptors of user preference in a series of linear regressions. We expected that the user would tend to like or dislike all activities in a hard cluster, since such activities would be similar to each other, and the user would like them to a similar extent. We further expected that a partial membership in a fuzzy cluster would correlate with the user preference for an activity if the user liked or disliked the underlying cluster. The results suggested that clustering could indeed be used to describe leisure preference. Hard clustering appeared to be more effective than fuzzy clustering, but both of them could be combined for an even more effective approach to preference modeling than using them separately. Finally, clustering could be combined with the four most effective of the 17 leisure dimensions to account for 27% of variance in the user preference for leisure. Such result suggests that a combination of leisure dimensions and cluster membership can be used as leisure activity features, which may be suitable for and useful in content-based and hybrid leisure recommendation systems.