Users’ tourism activities are often influenced by the visiting time of Point-of-interests (POIs). Considering the prevalence of location-based social networks(LBSNs) and user-generated- content (UGC), this study tries to propose a novel method to deduce the temporal characteristics for city POIs. To that end, first, we introduce the NLP method to extract tourism-related features for POIs from massive online documents. Then, we analyze the distribution of the tourists’ visiting time for a specific POI by taking each check-in made by an individual visitor as a preference voting. Finally, we propose a learning model based on matrix factorization to derive the temporal characteristics of each tourism feature, which is the fundamental for the model proposed in this work for predicting the feasible visiting time for a POI. Our experimental results on two real-world datasets show that the methods proposed in this work are efficient for discovering the feasible visiting time for city POIs.