Semantic Twitter: Analyzing Tweets for Real-Time Event Notification

Abstract

Twitter, a popular microblog service, has received much attention recently. An important characteristic of Twitter is its real-time nature. However, to date, integration of semantic processing and the real-time nature of Twitter has not been well studied. As described herein, we propose an event notification system that monitors tweet (Twitter messages) and delivers semantically relevant tweets if they meet a user’s information needs. As an example, we construct an earthquake prediction system targeting Japanese tweets. Because of numerous earthquakes in Japan and because of the vast number of Twitter users throughout the country, it is sometimes possible to detect an earthquake by monitoring tweets before an earthquake actually arrives. (An earthquake is transmitted through the earth’s crust at about 3–7 km/s. Consequently, a person has about 20 s before its arrival at a point that is 100 km distant.) Other examples are detection of rainbows in the sky, and detection of traffic jams in cities. We first prepare training data and apply a support vector machine to classify a tweet into positive and negative classes, which corresponds to the detection of a target event. Features for the classification are constructed using the keywords in a tweet, the number of words, the context of event words, and so on. In the evaluation, we demonstrate that every recent large earthquake has been detected by our system. Actually, notification is delivered much faster than the announcements broadcast by the Japan Meteorological Agency.