User Similarity from Linked Taxonomies: Subjective Assessments of Items
Makoto Nakatsuji, Yasuhiro Fujiwara, Toshio Uchiyama and Ko Fujimura
Subjective assessments (SAs) assigned by users against items, such as 'elegant' and 'gorgeous', are common in reviews/tags in many online-sites. However, previous studies fail to effectively use SAs for improving recommendations because there are few users who rate the same items with the same SAs, which triggers the sparsity problem in collaborative filtering. We propose a novel algorithm that links a taxonomy of items to a taxonomy of SAs to assess user interests in detail. It merges SAs assigned by users against an item into subjective classes (SCs) referring to a taxonomy of SAs prepared for each service domain. It also reflects SAs/SCs assigned to an item to the classes that includes that item. Thus, it can compute the similarity of users from not only SAs/SCs against the item but also those reflected to item classes, and thus overcome the sparsity problem. Our evaluation used data from a popular restaurant review site shows that our method generates more accurate recommendations than previous methods. Furthermore, we found that SAs assigned against items in a few classes, that is to say concrete SAs, are more useful in accurate recommendations than others.