In an online collaboration system such as Wikipedia, edit history is stored as revisions. Topics of articles or categories grow and fade over time, and evolutionary information is retained in its edit history. We consider that a great amount of information that is related to real life events is embedded in such edit history of documents. This paper focuses on a particular temporal text mining task: effectively extracting keyphrases from burst periods in the edit history of Wikipedia articles or category. We first combine the ARIMA model with a decay function to find typical edit burst periods, then perform keyphrase extraction on burst periods to reveal topics of bursts. However, keyphrase extraction methods, such as TextRank, do not consider temporal trends in text stream. In this paper, we propose TextRank_nfidf which reflects temporal trends into phrase node weights, by computing smoothed difference of editing frequency between revisions. We confirm that detected bursts and keyphrases are matching well with events along the timeline.