Yuan, Y., Gao, Y., Moen, H., Isometsä, E., Marttinen, P. and Aledavood, T.
Leveraging Large Language Models for Digital Phenotyping: Detecting Depressive State Changes for Patients with Depressive Episodes.
In review
-
This preprint explores how large language models (LLMs) can be used for digital phenotyping to detect changes in depression severity among patients with major depressive episodes. By analyzing behavioral data from smartphones and wearables, the study compares few-shot prompting and fine-tuning strategies (e.g., embedding-only vs. QLoRA) and finds that LLMs outperform traditional machine learning models. The study demonstrates the promise of LLMs in enabling more personalized, real-time mental health monitoring, but also underscores the need for careful clinical oversight, transparency, and ethical considerations to mitigate risks related to bias and limited model interpretability.
Yuan, Y., Kasson, E., Taylor, J., Cavazos-Rehg, P., De Choudhury, M. and Aledavood, T.
Examining the Gateway Hypothesis and Mapping Substance Use Pathways on Social Media: Machine Learning Approach
JMIR Form Res 2024;8:e54433
-
This project investigates the transitions between different levels of substance use using Reddit data. By analyzing over 2.29 million posts from 1.4 million users, the study uses machine learning to predict patterns of substance use escalation and de-escalation. Results show that specific linguistic cues can indicate these transitions, supporting aspects of the gateway hypothesis. The findings highlight the potential of using social media analysis for understanding substance use behaviors.
Yuan, Y., Saha, K., Keller, B., Isometsä, E.T. and Aledavood, T.
Mental Health Coping Stories on Social Media: A Causal-Inference Study of Papageno Effect
Proceedings of the ACM Web Conference 2023
-
The project investigates the positive role social media can play in preventing and mitigating suicidal ideation and behaviours, a phenomenon known as the Papageno effect. With the rise of social media platforms, individuals often share their experiences and struggles with mental health, providing a rich source of data for understanding the impact of these shared stories. They then measured the psychosocial shifts in affective, behavioural, and cognitive outcomes in longitudinal Twitter data before and after engaging with the coping stories. The study offers insights into social media as a tool for supporting mental well-being and provides insights into practical and platform design implications.
-
This project investigates the disproportional impacts of the COVID-19 LGBTQ+ online community. Due to pre-existing social disadvantages and health disparities for minorities, including the LGBTQ+ community, the COVID-19 pandemic has had a disproportionate impact on minorities. The project analyses minority stress, which is the unique pressure faced by the LGBTQ+ community because of their identities. The project found an increase in expressions of minority stress and anger and a decrease in positive emotional words during the pandemic. These findings highlight the need for targeted mental health support for the LGBTQ community during crises.
Yuan, Y., Verma, G., Keller, B. and Aledavood, T.
Minority Stress Experienced by LGBTQ Online Communities during the COVID-19 Pandemic
Proceedings of the International AAAI Conference on Web and Social Media. Vol. 17. 2023.