Item recommendation in social tagging systems using tag network

How to profile users and items is a key problem for recommendation in tagging systems. In contrast to tag vector based methods which ignore the semantic relations between tags, we present a novel profiling method based on a weighted tag network model to fully exploit the rich tag relations. Furthermore, by considering the extent of other users' usage of tags, we present a novel NTF-IUF-IIF method to calculate weights for tags, which can seize the user's preference accurately. Instead of a single document of traditional methods, it is the first effort to regard each user as a document collection, which enables the statistics of all items. Then the extent of other users' usage of tags can be counted via the global item information, and then used as a factor for accurate tag weighting. Finally, a Fusion Method (FM) is proposed for measuring similarities between tag networks of users and items to get the recommendation lists. Experimental results on MovieLens and CiteULike datasets validate the effectiveness of our methods.