A Fundamental Attempt to Extract Unusual Geo-Tagged Tweets

Abstract

Twitter has been recognized as a popular communication tool. In this paper, we report a fundamental attempt to extract personal tweets on unusual and uncommon events. Because unusual tweets represent exceptional situations and/or strongly desired ones, by using such tweets, we can expect to build a sophisticated user pro le used in further information services such as recommendation systems. Among several aspects on uncommonness of tweets, we focus on contents in tweets as well as locations where the tweets are posted, and prepare several evaluation criteria of unusualness of tweets. A preliminary experiment using small real datasets is conducted to assess the feasibility of the proposed framework.