Stops along taxi trajectories, such as picking up and dropping off passengers, are spatially clustered and related to certain attributes of places where stops are made. To detect the hidden knowledge regarding these places, this article examines the semantics of massive taxi stops in a large city. Each taxi trajectory is modeled as a series of sequential semantic stops labeled by street names. All the trajectories can be examined as a document corpus, from which the hidden themes of the stops are identified through Latent Dirichlet Allocation model. Conventional GIS tools are coupled with topic modeling toolkit to visualize and analyze potential information of stop topics for understanding intra-city dynamics. The effectiveness of this approach is illustrated by a case study using a large dataset of taxi trajectories including approximately 4,000 taxis in Wuhan, China.@en