Recommending Activities for Mental Health and Well-Being: Insights From Two User Studies

Engaging in daily activities that engender positive affect (e.g., exercise and socializing) is critical for emotional well-being and is effective in reducing clinical depression. However, current digital mental health interventions have not exploited recommender approaches to encourage such healthy behaviors. This paper tests the feasibility of recommending personalized, healthy activities to users. Using two mobile applications, we collected high-quality data about specific healthy activities from two populations: a clinical sample diagnosed with a mood disorder (n = 318 activities/user) and a non-clinical sample (n = 59 activities/user). Activities were labeled with a type (e.g., social, leisure, work) and rated for their impact on mood. We used a probabilistic Naive Bayes (NB) Classifier and a Support Vector Machine (SVM) to model the activities as a bag-of-words to predict mood outcomes. We separate the analysis into a generalized model where we pooled all participants, comparing it with a personalized model. In both the clinical and non-clinical samples, there was a significant difference between the models. Both NB and SVM favored the personalized model after collecting 58.92 (SD = 20.96) activities. This research sheds light on recommendations for mental health, showing that personalization is key for recommending the right activity to each user.