Tagging is one of the most important ways to categorize or indexing information at the age of Web 2.0. To handle the disadvantages of tagging systems such as inconsistentcies, redundancy and incompleteness, tag recommendation methods improve the quality of tags by providing candidate tags. In order to further improve the quality of tag recommendations, a tag recommendation method is proposed which bases on a combined analysis of the relations of objects in a tagging system and the content of resources. An LDA based generative tagging system model TSM/Forc that models object relation and resource content in a combined way is introduced, together with a probabilistic based tag recommendation method and a Gibbs sampling based model parameter estimation approach. Experiment results show that the proposed method could provide more accurate recommendations than the latest methods.