In this paper

we study the language features across depression of twitter users

and show that depression differ from ordinary users in terms of linguistic traits. Depressions tend to post less tweets

write shorter posts

and use more pronouns. In addition

a novel model empath was introduced in analyzing the language of depression

and it obtain more language cues for depression. These findings were applied to depression prediction tasks

the result show that Random Forest classifier has the best performance and the feature based on Empath has the largest contribution in classifications.