In this paper, we introduce a novel topic extraction method. Firstly, we divide a video into events based on target character's appearance and disappearance. Specifically, each event is an interval where the target character performs a certain action. And, it is characterized by a specific pattern of shots where the character appears and shots where he/she disappears. Then, we define a "topic'' as an event where the target character performs an interesting action (e.g. fight, chase, kiss and so on). We extract such topics as events containing abnormal patterns, called "bursts''. The experiments on different videos validate that character's appearance and disappearance are effective for obtaining semantically meaningful events. From these events, we could extract many interesting topics.