8 , 2016-03-24 , 法政大学大学院理工学・工学研究科
Recently Microblogs such as Twitter become popular. The application treats short messages in real time. Users can indicate whether the messages are categorized by attaching so called ‘hash-tag’ according to their interests. When a certain social event such as sport game is happening, the sequence of message with a group of ‘hash-tag’ describes the discourse of the event. Once a special incident such as shooting in a soccer game occurs, sequence of tweets shows burst in time-line. In this research, in order to understand the situation in large social event, both burst phenomenon of messages and words appears in messages are utilized. Algorithm named ‘aggregation pyramid’ is used for detecting group of messages in an event and analysis based on ‘tf-idf’ for extracting words to describe the situation when a burst is happening. To evaluate proposed automatic extraction of specific incident in a soccer game, human summary and the results of machine extraction is compared. The results shows good matches and availability of proposed method.