Introduction
Understanding how individuals' behaviour 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). 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.
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 anonymising 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 allowing only certified access.
Project aims
The major goal of this project is to develop personalised longitudinal sensors from individuals' language use and user-generated content (UGC) to better understand changes in behaviour over time, with applications in mental health.
To achieve this goal, the project has the following objectives:
- Create underlying representations of an individual through their language and other content they generate.
- Overcome data sparsity and privacy issues by generating synthetic meaningful data for longitudinal mental health tasks.
- Create models for understanding and predicting individual behaviour over time by fusing asynchronous and heterogeneous data.
- Develop methods for understanding the behaviour baselines of individuals and changes in these over time.
- Create an evaluation framework of methods in the real world.
- Create summaries of individual behaviour over time for clinicians and individuals.
- Co-design new instruments and measures to support diagnosis, monitoring and caring in mental health.
- 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.
Applications
The project has a range of different stakeholders:
- Clinicians
Clinical experts in mental health and dementia would be able to use the new instruments for measuring changes in mood and linguistic ability as well as the interpretable multi-input summaries that the project will create to help them make more informed diagnoses and assessments. - Online platforms
Interaction with the project's industrial collaborators (Facebook, Mumsnet, TalkLife) can help them improve the quality of their services provided to the public by more efficient and effective monitoring of user posts and identification of individuals at risk. - Wellness industry
By releasing a software library for time-sensitive sensors from language and heterogeneous user-generated content as well as a real-world evaluation framework, the project will allow developers in the wellness industry to create truly personalised applications that consider the trajectory of individuals over time, thus increasing the usefulness of such applications. - Non-profit organisations for mental health
Non-profit organisations for mental health in the UK (e.g. Mental Health Foundation), and internationally (e.g., Mind), could use the time-sensitive sensors from user-generated content, the interpretable summaries and the new instruments to better support people going through mental health difficulties. - Individuals
Private individuals would be able to install a version of the software produced on their laptops or phones to monitor their own changes in mood and language ability over time, thus enabling them to better understand and manage their own mental well-being, collect their own data and be more in control of their own health.
Organisers
Researchers and collaborators
Adam Tsakalidis
Visiting Researcher
