Often Hyperconnected, Seldom understood

Lead Research Organisation: University of Reading
Department Name: Computer Science

Abstract

Motivation and Overview
The increasing proliferation of social media has led to a number of positive effects in supporting social engagement and cohesion as well as sharing insights, and, for some, a sense of belonging to same (sub) group or community of interest. The prevalence and ubiquity of social media linking as an "always-on" lifestyle whilst serving as a way to feel online togetherness, can paradoxically also become a route to feel loneliest and misunderstood due to various behaviours of participants such as insensitive , aggressive, deceptive and obsessive conduct meted out to the most vulnerable.

Over the last decade much work has focused on analysis of patterns of behaviour and self-expression on the internet and its possible correlates including the increasing trend in mental illness, particularly amongst the adolescent and young adults as heavy users of social media.

According to the World Health Organisation "Half of all mental illness begins by the age of 14, but most cases go undetected and untreated. In terms of the burden of the disease among adolescents, depression is the third leading cause, and, suicide is the second leading cause of death among 15-29-year-olds" https://www.who.int/mental_health/world-mental-health-day/2018/en/

The proposed research study aim to address the analysis of undetected mental and/or physical illness based on the social media behaviours of users; particularly the lonely and the vulnerable who for some reason may be unable to articulate their real feelings in the physical or virtual space.
However their patterns of presence, absence, participation, self-expression and verbal or written exchanges with others may help reveal physical or mental ill-health which could otherwise go undetected and may indeed be serious and need urgent attention.

Thus the thesis contends that using machine learning and analytics techniques it should be possible to detect changes in the behaviour of the online persona of a user, as exhibited in terms of the above patterns, that may correlate with the onset or deterioration of mental or physical health such as suicidal thoughts, clinical depression, anxiety and obsessive disorders, or indeed the onset of sever life threatening infectious diseases such as meningitis, septicaemia etc.

Data Assets & Ethics
In the first instance an ethically-guided approach to data assets engineering will ensure the acquisition of open data sources with data-subject identity anonymised at source. Over the life-cycle of the project it is possible that some synthetic data will be generated, based on established behavioural profiles for various personality disorders, for model building and testing purposes.

Methodology
This PhD study demands challenging and innovative deployment of various AI and Data Analytics techniques notably Metadata, Ontology and Privacy Engineering. Natural Language Processing, Spatio-Temporal Data Stream Mining, Uncertainty Modelling and Generative Networks.
This study is funded through a studentship grant from EPSRC and the School of Mathematical, Statistical and Computational Sciences, under EPSRC Contract No

Publications

10 25 50

Studentship Projects

Project Reference Relationship Related To Start End Student Name
EP/R513301/1 30/09/2018 30/08/2025
2271816 Studentship EP/R513301/1 30/09/2019 29/09/2022 Rhian Taylor