Monitoring Students' Well-Being through Journal Analysis – Study Protocol of an Explorative Approach using Natural Language Processing on Typed and Transcribed Entries to Monitor and Generate Personalized Feedback

Abstract

Background: More than one-third of German students report high emotional exhaustion. One effective and low-cost method for improving emotional well-being is journaling, and advances in artificial intelligence (AI), particularly in Natural Language Processing (NLP), enable automated analysis of journal content for mental health monitoring. These analyses can be returned to users as personalized feedback, thereby enhancing the positive effects of journaling through increased self-reflection and creating an incentive for continuous use, which, in turn, improves monitoring. Journaling apps offer various input modalities (e.g., typing, speaking), potentially further increasing participation. Despite the promising potential of AI-powered journaling apps, their effects on mental health have so far been scarcely investigated. Research Question: This study investigates the performance of NLP models in predicting emotional well-being from journal entries. Specifically, it examines whether typed or spoken entries provide a more suitable input modality for these predictive models. Additionally, we explore whether receiving feedback is positively evaluated and which types of feedback students prefer. Method: In a two-week observational study, N = 100 university students (aged 18 years and older) will be recruited and randomly assigned to one of two groups (speaking vs. typing). Following a baseline assessment, participants will submit daily journal entries and annotate them using reflective questionnaires on emotional well-being, stress, and journal topics. This information will be presented to participants as personalized feedback. Additionally, the performance of NLP models in predicting emotional well-being from journal entries will be evaluated separately for the two groups. Expected Results: We expect that emotional well-being can be predicted from journal entries using NLP, with transcribed entries yielding higher accuracy than typed entries. We also expect the feedback to be well-received. Discussion: This study explores an accessible and engaging journaling application for monitoring and providing feedback on emotional well-being. It addresses several challenges in e-health research (e.g., high dropout rates, low user engagement). If this innovative approach yields strong user engagement and predictive performance, future work should evaluate automatically generated NLP-based feedback in journaling interventions to promote mental health. Trial registration: This trial was registered in the DRKS register (DRKS-ID: DRKS00034660) in July 2024 (16.07.2024).