Event Duration Detection on Microblogging

• Main Authors: Yi-Shiang Tzeng, 曾奕翔
• Other Authors: 鄭卜壬
• Format: Others
• Language: en_US
• Published: 2013
• Online Access: http://ndltd.ncl.edu.tw/handle/65613387397704895432

碩士 === 國立臺灣大學 === 資訊網路與多媒體研究所 === 101 === Twitter, a popular microblogging service, has passed the 500 million users. If we consider widely distributed Twitter users as social sensors, the sensor network provides us a snapshot of the real world. In this study, we take the advantage of Twitter data to detect events and model their durations. We choose rain as our target event, and build an on-line weather station to tell whether it rains or not for any given location and time. Our system contains two stages. In the first stage, we find out truly rain-related tweets from candidate pool to deal with the inherent noises in Twitter. In the second stage we construct an aging based model to simulate the life cycles of rain events. We compare our model to other event detection based methods. Our results show that it’s not feasible to transform the problem of detecting duration of rain events to multiple rain events detection problems. We further figure out how spatiotemporal factors and the properties of events influence our model. User behaviour is also carefully discussed. Finally, we extend the rain event detection system to rain forecast system.