Enhancing personalized recommendations on weighted social tagging networks

Recently, social tagging systems have been widely applied in web systems and some physical properties have been found applications in efficiently and effectively personalized recommendation. Social tags can provide highly abstract information about not only item contents but also personalized preferences, hence they might help generate better personalized recommendations. However, how to find out the relevant yet diverse items that are not associated with any tag remains an open question for us. In this paper, we assume a basic attraction may exist for each item. Moreover, considering both personal and global vocabulary, as well as such attractor, we apply diffusion-based recommendation algorithm in weighted social tagging networks. We then evaluate it in a real-world data set Del.icio.us. Experimental results demonstrate that the usage of both tag information and attractor can significantly improve diversity of personalized recommendations, and thus it can be regarded as an alternative recommendation method.