Performance of Content-Based Features to Detect Depression Tendencies in Different Text Lengths

Text analytics have been widely used nowadays in the field of mental health to predict onset mental health issues such as depression and anxiety, with the intention to perform early intervention. Most existing works focuses on looking at how such mental health issues can be predicted based on social media data. These texts are often short and straightforward as compared to blogs and journals. In this paper, we are interested in comparing the performance of a classification model in classifying long texts and short texts as having depression tendencies. An existing model that can perform well in classifying short texts using content-based features was adopted and tested on longer texts. From the result, it is found that compared to shorter text, content-based features performed worst in long texts whereby all five classifiers used produced an accuracy of less than 0.65.