Prediction of Depressed Users' Emotional States in Online Communities Using Emotion Transition Matrices

In this paper, we developed a new emotion prediction method based on Markov modeling for predicting the direction of emotional state shifts in depressed social media users. This improves the accuracy of detecting mood fluctuations and offers an early warning capability that is critical for clinical intervention. We developed a "first-order mood shift enhancement prediction matrix" (FST-EPM) and a "second-order mood shift enhancement prediction matrix" (SST-EPM) using historical mood data posted by depressed users in the Weibo depression community to better track and interpret depressive mood changes for early detection. In our experiments, we compared them based on the prediction accuracy, recall, and F1 score metrics. The results indicate that the proposed method outperforms traditional machine learning models by a wide margin in validity and accuracy, with its predicted values closer to actual observations.