Understanding mood changes of individuals with depressive disorders is crucial in order to guide personalized therapeutic interventions. Based on diary data, in which clients of an online depression treatment report their activities as free text, we categorize these activities and predict the mood level of clients. We
apply a bag-of-words text-mining approach for activity categorization and explore recurrent neuronal networks to support this task. Using the identified activities, we develop partial ordered logit models with varying levels of heterogeneity among clients to predict their mood. We estimate the parameters of these models by employing Markov Chain Monte Carlo techniques and compare the models regarding their predictive performance. Therefore, by combining text-mining and Bayesian estimation techniques, we apply a two-stage analysis approach in order to reveal relationships between various activity categories and the individual mood level. Our findings indicate that the mood level is influenced negatively when participants report about sickness or rumination. Social activities have a positive influence on the mood.
By understanding the influences of daily activities on the individual mood level, we hope to improve the efficacy of online behavior therapy, provide support in the context of clinical decision-making, and contribute to the development of personalized interventions.