What is the author like?

Word-of-mouth marketing on microblogs (e.g. Twitter) has become more important given the increasing number of users and posts. However, unlike questionnaire surveys, there is a problem in that most users do not specify their attributes (gender, age, etc.) in their profiles. We resolve this issue by employing user attribute estimation. Existing methods use bag-of-words features extracted from manually labeled users' posts to estimate user attributes. We propose a new method that focuses on the microblog structure and the topics of words in the user's posts. This method is very accurate in estimating the major marketing attributes; e.g. it is about 90% accurate for gender. Our method is designed for marketing use, but we intend to enhance it so that it can estimate various and more segmented attributes for personalization use.