Incident Related Tweet Extraction with Density Ratio Estimation

These days social media services are widespread and are infrastructure of communication in the Internet. Since Twitter is one of the most famous real-time communication services, we can grasp opinions of crowds in the real world analyzing tweets. There are usually various kind of opinions in Twitter and you need to deal with the opinions carefully. In this paper we focus on tweets on an incident and extract tweets reflecting sufferers’ opinions. When a incident happens, vast amount of tweets are created by many Twitter users. We compare tweets by sufferers with ones by others and extract tweets unique to the sufferers with density ratio estimation. In experiments we confirm that our proposed method can extract tweets including sufferers’ opinions.