TV drama has various genres such as "romantic" and "crime." However, recent TV dramas cannot be categorized into a single genre, and features of a variety of genres sometimes coexist. In this study, we analyze the user's reaction according to the progress of a TV drama from live commentary tweets and clarify the features of the scene. A "development pattern" is extracted based on the characteristics of scenes, and a clustering of TV dramas is performed, which is different from clustering based on genre. The features of each scene in dramas is modeled by a topic model. The topic distribution of a scene is extracted using LDA as one document by combining the live comments posted during one scene. The "development pattern" of a drama is represented as the wave of the topic connecting the topic distribution of each scene.