Revealing Runners’ Mental States And Behavioral Strategies Through Natural Language Processing And Wearable Devices

PURPOSE: To qualitatively characterize the emotional landscape of amateur runners and discover relationships between mental wellbeing and running behavior. METHODS: Self-reported pain level (4-point scale) and mental wellbeing (5-point scale, Terrible to Great) ratings as well as wearable device data were collected from runners over a period of 12 weeks. Participating runners also provided open-ended text comments explaining their mental wellbeing rating choice each week. Using the BERTopic framework, we conducted guided neural topic modeling with emotion-related seed words on the mental wellbeing explanations to identify common themes and affective states experienced by the runners. We divided participants into three groups based on their most frequent mental wellbeing ratings: Positive Mind, Neutral Mind, and Negative Mind, and engineered prompts for a hybrid SSM-transformer language model (Jamba-Instruct) to analyze the open-ended comments from these groups. To identify concrete relationships between mental wellbeing and overt running behaviors, we utilized the one-tailed Mann-Whitney U test to compare the distribution of total weekly mileage for each mental wellbeing rating with that of the next highest rating. RESULTS: 16,188 self-reported comments explaining the weekly mental wellbeing ratings of 1,847 runners (826 Male / 978 Female / 43 Other, Median Age = 42 years [IQR = 13.125 years]) were included in the natural language processing analysis. Topic modeling revealed distinct clusters of commonly co-occurring affective states. From the language model analysis, we extracted psychological traits and behavioral strategies that distinguished Positive Mind from Negative Mind runners, suggesting targets for behavioral interventions. Specifically, Positive Mind runners expressed connection to a concrete training goal and exhibited mindfulness and adaptability. For each mental wellbeing rating except 4/Good (U = 2,384,183, p = 0.054), runners’ total weekly mileage was significantly less than in weeks with the next highest rating (maximum p = 0.009) at a level of 0.05. CONCLUSIONS: Amateur runners experience a diverse array of mental states that can impact their training. Thematic analysis of these experiences suggests strategies for runners to increase their resilience.