Research on the Fluctuation Characteristics of Social Media Message Sentiment with Time Before and During the COVID-19 Epidemic

“Tree hole” refers to a social media formed after the death of a social media user, in which other users continue to leave messages due to emotional resonance. This paper focuses on exploring the fluctuation of emotions with time in a “tree hole” of social media such as Microblog, and provides ideas and support for suicide warning, rescue, and user portraits of patients with depression in the “tree hole”. In this paper, the dataset of 2,356,066 messages captured from the “tree hole” Microblog with the “tree hole” agent (i.e., an AI program) and pre-processed. Subsequently, the effective dataset was labeled by a text sentiment analysis model based on BERT and BiLSTM, and accordingly the sentiment was scored. Then the scored data was visualized and analyzed in the time dimension. Finally, it was found that the sentiment of the “tree hole” messages reached a trough at 4:00 am and a peak around 8:00 am. In addition, the overall trend of “tree hole” sentiment has fluctuated downwards from Monday to Sunday. We have concluded that the sentiment of patients with depression fluctuates regularly at some special time points, and special events such as the outbreak of COVID-19 and so on, have a great impact on the emotions of patients with depression. Therefore, it is necessary to strengthen warning and intervention for those who has expressed thoughts of suicide at special points to prevent the spread and fermentation of suicidal emotions in the “tree hole” in time. In addition, the rescue volunteers for patients with depression as Tree Hole Rescue Team should make corresponding adjustments to the rescue strategy when special events occur. This research is of great significance for the emergency response of “tree hole” depressed users in major events such as COVID-19 epidemic