[4Yin2] Interactive session 2

[4Yin2-08] Avatar’s social rhythm reflects its users’ depression.

〇Kenji Yokotani¹, Masanori Takano² (1.Tokushima University, 2.CyberAgent, Inc.)

Keywords:Depression, social rhythm, time series classification, avatar

Avatar's behavior is related to its users’ mental state. Based on this principle, we estimate their depressive levels from their avatar's behavior log. Participants in the first and second wave were 3361 and 658 Pigg Party users. Their responses to the depression questionnaire and their behavior logs in Pigg Party during the first and second waves were the current datasets. The dataset from the first wave showed that non-depressed users were more likely to show their avatar’s behaviors in a 24-hour cycle than depressed users. Further, we trained offence time predictor on the first wave data and predicted the second wave data’s depression levels with the predictor, resulting in an accuracy of 0.7562. Yet, other well-known predictors did not show such an accuracy. These results suggest that a machine learning model with medium complexity is desirable for estimating users’ depression based on their avatar’s behavior logs.