Abstract

Detecting clinical depression is an important task to find affected patients for effective treatment, especially in an early state with a higher effective treatment. This work proposes a method for automated detecting the possible depression-positive person from Facebook data, which refers to the user’s textual posts and usage behavior. A machine-learning classification then uses the data to create a model to determine features signifying depression-positive users. We consider used words and statistical data of actions made on Facebook platforms, such as the number of posts, comments, and replies a user made daily, along with time and frequency information of these actions. An experiment was conducted to examine the potential and capability of the proposed method. A model from Neural Networks’ behavior data yielded the best result, a 1.0 F1 score. In contrast, the model of text data from Neural Networks acquired the results as 0.88 F1 scores for classification results. From the models, we also obtain a list of significant features indicating a depression-positive state of users as keywords from text data and notable behavior from action data based on the calculated weight from machine learning.