Inferring user profile using textual content on twitter

Abstract

In the past few years online social media have risen as a key venue for communicating with the public and monitoring public opinions. People talk about movies they watch, restaurants they visit, and views they enjoy, insinuating their whereabouts. In order to weigh in the public opinions expressed on such social media as much as traditional poll results or to optimize businesses for specific class of users, the representativeness of the opinions has to be accounted for. A profile such as age, gender, and location of users is one of the key factors in the representativeness, but are not available by default in online social networking platform. The number of users who make their profiles public is relatively small, compared to the huge number of users in online social networking services and social media platforms. Besides, there are several studies inferring user profile on various social networking services, but none of them apply their methods on Korean Twitter users. In this work we propose a new framework to infer a Korean user`s main location of activities, age, and gender in Twitter using their textual contents. Our approach is based on a probabilistic generative model that filters local words, employs data binning for scalability, and applies a map projection technique for performance in inferring user’s main location. Also, we use classifier for inferring user’s age and gender and apply feature selection for filtering relevant features to classes. We evaluate our method with users who have focused GPS-tagged tweets or with manually annotated users who use profile-relevant words in their description data. For inferring Korean user’s location, we report that 60% of users are identified within 10km of their locations, a significant improvement over existing approaches. And for inferring user’s age and gender, we report that 75% and 88% of users are correctly identified.