Subjectivity detection based on multi-feature fusion

Subjectivity detection is a task of separating subjective contents from objective contents in text. While the detection itself is more than objective, the task is harder than topic classification and has drawn much focus recently. This paper proposed a multi-feature based method. Firstly, all sorts of candidate features are collected and then put into an improved SIMBA feature selector. In the last, the selected features are used as clues for a standard classifier. Experiment shows that the performance is not only higher than that of using standard classifier solely, but also higher than that of previous method on the same dataset.