Resource recommendation via user tagging behavior analysis

Tag-based resource recommendation is an interesting and important research topic and has been applied to a wide range of applications. The user’s tagging behavior usually reflects his/her interests in social tagging systems, however most existing work can not fully consider the features of user’s tagging behavior, such as tag frequency, time and ordinal position in tag assignments. In this paper, we employ the combination of cluster analysis and data fitting for extracting the correlations between user interests and the three features, and then present a novel user interest model based on the features to compute the user interest degree. In addition, we propose a collaborative filtering based approach, in which top-k similar users are filtered by resource-interest-based profiles; resource similarities are obtained by tag-frequency-based profiles; the candidate resources are then ranked according to the user interest model, resource profile similarity and user profile similarity. The experiment results conducted on two real-world datasets demonstrate that the proposed approach outperforms the traditional collaborative filtering baselines.