Instance-based collaborative filtering with fuzzy labels.
Eyke Hüllermeier · 2003
In recommender systems, user ratings of items are often represented in terms of linguistic labels such as “fair ” or “very good”. We investigate the potential of fuzzy sets as a means for modeling such labels, especially with regard to collaborative filtering algorithms. We propose a related fuzzy version of instance-based (memory-based) collaborative filtering and argue that it leads to more informative and accurate predictions. These claims are validated by means of two experimental studies.