Location context in social media plays an important role in many applications. In addition to explicit location sharing via popular check in service, user-posted content could also implicitly reveals users location context. Identifying such a location context based on content is an interesting problem because it is not only important in inferring social ties between people, but also vital for applications such as user profiling and targeted advertising. In this paper, we study the problem of location type classification using tweet content. We extend probabilistic text classification models to incorporate temporal features and user history information in terms of probabilistic priors. Experimental results show that our extensions can boost classification accuracy effectively.