The objective of collaborative filtering is to cluster the consumers based on their past behavior using similarities to make recommendations. The rating is not a target variable in this case, but rather an input variable. So this is a clustering / unsupervised learning technique. Hope this clarifies the confusion!