User-Characteristics Topic Model

This paper proposes a method to capture user’s characteristics in a topic model frame, where user characteristics act as a latent variable that does not depend on texts. As it is obvious that different people possess different characteristics, they may perform differently even when they are facing the same document. These different characteristics can be showed as different views or different wording preference. We think this phenomenon has a great impact on modeling texts written or labelled by different people, especially on topic modeling. Experiments show that the model with user characteristics outperforms the original models and other similar topic models on corresponding tasks. A combination of the user’s characteristics can not only provide better performance on normal topic modeling tasks, but also discover the user’s characteristics.