Trend Extraction Method using Co-occurrence Patterns from Tweets

Abstract

We can feel free to post the information such as personal events using Twitter one of the popular micro-blogging service. However, the collection of information is limited by the human power only, therefore, the method of collecting trends automatically is important. Existing web services focus on the number of tweets for getting trends. However, a time lag was occurred for extracting the trends. In this paper, we propose the trend extraction system for twitter in real time by paying attention to the co-occurrence patterns. Our system can learn the new key patterns at the same time not only using the picked up trend biterms, previously. Furthermore, we evaluate the efficiency of the proposed method of extracting the trends from twitter by the comparative experiments. We demonstrate that our proposed method can extract accurately and widely without time-lags compared with the existing service (Realtime Yahoo Search).