Item-based collaborative ranking.

Neighbor-based collaborative ranking algorithms exploit users’ pairwise preferences to predict how they will rank items. Current neighbor-based algorithms lie in the category of user-based recommendation methods: they calculate users’ similarities over pairwise preferences, estimate the unknown pairwise preferences, and finally, infer the ranking of items for the target user. However, it still is an open question how to adapt item-based recommendation for neighbor-based collaborative ranking. The more specific question is how to calculate items’ similarities in a pairwise preference dataset, and how these similarities can be employed to infer the total ranking of items for the target user. This paper presents a novel recommendation approach, PreNIt, that exploits preference networks for item-based collaborative ranking. PreNIt models the users’ pairwise preferences as two novel bipartite networks with labeled edges. These labeled edges enable us to model the choice context in which items are preferred/not preferred by the user. Once the networks are constructed, PreNIt finds the transitive similarities of items using a new personalized ranking algorithm in graphs with labeled edges. Experimental results shows the significant outperformance of PreNIt over the state-of-the-art algorithms. [ABSTRACT FROM AUTHOR]