Difference between revisions of "Moments of Change in Mood (Switch/Escalation)"

Revision as of 21:10, 6 March 2022

Introduction

Understanding how individuals' behavior changes over time is very important, particularly in monitoring mental health conditions, which affect one in four people globally. This project will create AI models and natural language processing methods that are able to track the progress of an individual over time, based on their language use and other interactions through the use of digital media, called 'user-generated content' (UGC) [1][2]. Major outcomes include software implementing time-sensitive sensors from heterogeneous UGC, and new tools for diagnosis and monitoring of mental health conditions, in keeping with good clinical practice [3].

This project involves working with sensitive user-generated linguistic and heterogeneous content, the protection of which is a top priority of the work. The research protects the users first by anonymizing all of their data that are used during the project and then by storing them in a secure environment that ensures their safety, non-transferability, and allows only certified access.

Project Aims

The major goal of this project is to develop personalized longitudinal sensors from individuals' language use and user-generated content (UGC) to better understand changes in behavior over time, with applications in mental health. We have defined moments of change and timelines for individuals based on moments of change. Moments of change are signaled by changes in mood, symptoms, life events, aha moments, or therapist interventions. Detection of the moments of change can facilitate the identification of switch in mood or escalation in mood, which can help bracket the text region that most likely contributes to increasing the severity of risk of self-harm or other mental ill-health conditions [8]. In a recent study [4], authors differentiated between time-variant and time-invariant assessment of suicide risk using a clinically-authorized questionnaire and found that longitudinal study effect in early detection of transitions from low-risk to high-risk level Dataset. Further [5] and [6] have been working with data from the TalkLife peer support network, a corpus of a clinical therapy session, and a corpus of longitudinal language and multi-modal data, respectively. Moreover, a system that precisely detects moments of change, can help create a dynamic peer support network, like the one shown in a recent study on Reddit [7].

To achieve this goal, the project has the following objectives:

1. Create underlying representations of an individual through their language and other content they generate.
2. Overcome data sparsity and privacy issues by generating synthetic meaningful data for longitudinal mental health tasks.
3. Create models for understanding and predicting individual behavior over time by fusing asynchronous and heterogeneous data.
4. Develop methods for understanding the behavior baselines of individuals and changes in these over time.
5. Create an evaluation framework of methods in the real world.
6. Create summaries of individual behavior over time for clinicians and individuals.
7. Co-design new instruments and measures to support diagnosis, monitoring, and caring in mental health.
8. Create new software libraries to support all of the above.

Finally, while the major focus of this work is around mental health, the aim is to develop models that can be incorporated in other domains leveraging UGC within the healthcare domain and beyond. The project has a range of different stakeholders: Clinicians, Online platforms, the Wellness industry, Non-profit organizations for mental health, and Individuals.

Funding

• The project is funded through an EPSRC-UKRI grant to facilitate US-UK collaboration.
• Timeline: 11/01/2021 – 04/01/2022
• Part Award Amount to AIISC: $23,928

People

Lead Contributors: Maria Liakata, Adam Tsakalidis, Manas Gaur, Federico Nanni

Organizational Support and Advising: Philip Resnik, Dana Atzil-Slonim, Ayah Zirikly

References

1. Yazdavar, Amir Hossein, Mohammad Saeid Mahdavinejad, Goonmeet Bajaj, William Romine, Amit Sheth, Amir Hassan Monadjemi, Krishnaprasad Thirunarayan et al. "Multimodal mental health analysis in social media." Plos one 15, no. 4 (2020): e0226248.
2. Tsakalidis, A. and Liakata, M., 2020, November. Sequential modelling of the evolution of word representations for semantic change detection. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP) (pp. 8485-8497).
3. Gaur, Manas, Amanuel Alambo, Joy Prakash Sain, Ugur Kursuncu, Krishnaprasad Thirunarayan, Ramakanth Kavuluru, Amit Sheth, Randy Welton, and Jyotishman Pathak. "Knowledge-aware assessment of severity of suicide risk for early intervention." In The world wide web conference, pp. 514-525. 2019.
4. Gaur, Manas, Vamsi Aribandi, Amanuel Alambo, Ugur Kursuncu, Krishnaprasad Thirunarayan, Jonathan Beich, Jyotishman Pathak, and Amit Sheth. "Characterization of time-variant and time-invariant assessment of suicidality on Reddit using C-SSRS." PLoS ONE 16, no. 5 (2021).
5. Tsakalidis, A., Atzil-Slonim, D., Polakovski, A., Shapira, N., Tuval-Mashiach, R. and Liakata, M., 2021, June. Automatic Identification of Ruptures in Transcribed Psychotherapy Sessions. In Proceedings of the Seventh Workshop on Computational Linguistics and Clinical Psychology: Improving Access (pp. 122-128).
6. Gkoumas, D., Wang, B., Tsakalidis, A., Wolters, M., Zubiaga, A., Purver, M. and Liakata, M., 2021. A Longitudinal Multi-modal Dataset for Dementia Monitoring and Diagnosis. arXiv preprint arXiv:2109.01537.
7. Gaur, Manas, Kaushik Roy, Aditya Sharma, Biplav Srivastava, and Amit Sheth. "“Who can help me?”: Knowledge Infused Matching of Support Seekers and Support Providers during COVID-19 on Reddit." In 2021 IEEE 9th International Conference on Healthcare Informatics (ICHI), pp. 265-269. IEEE, 2021.
8. Rami Sawhney, Atula Tejaswi Neerkaje, Manas Gaur, A Risk-Averse Mechanism for Suicidality Assessment on Social Media, In ACL 2022 Main Conference (to be published after the conference)