Many tourists upload content about tourist attractions to social media sites. Understanding tourist mobility using the contents benefits numerous applications, such as tourism recommendation and city planning. In this study, we propose a method for trajectory clustering of user movement. Our approach uses an improved skip-gram model to learn movements between a pair of locations quantized by their latitude and longitude. The generated embedding vectors represent the relationships between the movements from one area to the next. We demonstrated that the embedded vectors generated using our proposed method could cluster users' trajectories.