Person, Organization, or Personage: Towards User Account Type Prediction in Microblogs

During the past decade, microblog services have been extensively utilized by millions of business and private users as one of the most powerful information broadcasting tools. For example, Twitter attracted many social science researchers due to its high popularity, constrained format of thought expression, and the ability to reflect actual trends. However, unstructured data from microblogs often suffer from the lack of representativeness due to the tremendous amount of noise. Such noise is often introduced by the activity of organizational and fake user accounts that may not be useful in many application domains. Aiming to tackle the information filtering problem, in this paper, we classify Twitter accounts into three categories: “Personal”, “Organization” and “Personage”. Specifically, we utilize various text-based data representation approaches to extract features for our proposed microblog account type prediction framework “POP-MAP”. To study the problem at a crosslanguage level, we harvested and learned from a multi-lingual Twitter dataset, which allows us to achieve better classification performance, as compared to various state-of-the-art baselines.