Using data analysis and machine learning to derive insights from text-based mental health and wellbeing digital media

Abstract

The thesis will examine the use of data analysis and machine learning (ML), to derive insights from text-based mental health and wellbeing data. To understand current research in this area, a systematic review was performed, and the findings identified that machine learning and data analysis techniques offer valuable insights into mental health. Analysis was then carried out on three mental health and wellbeing datasets. Two studies were carried out on the first dataset which was a collection of happy moments, to examine how happiness might vary across demographic groups and explore if ML classification accuracy changes when applied to these groups. The next dataset analysed included responses from an online survey and was analysed to discover the attitudes relating to the use of chatbots among mental healthcare professionals. The final dataset contained media articles that did not adhere to suicide reporting guidelines and was analysed to discover patterns in relation to non-adherence. The contributions to knowledge reveal that happiness varies across demographic groups, with higher classification accuracy for married and parent subgroups. Mental healthcare professionals showed positive attitudes towards chatbots, especially those with more experience. Analysis of media articles linked suicide methods to location details and highlighted a concerning lack of helpline information, particularly in celebrity suicide reports, which is also concerning given the link between sensational reporting in relation to celebrity suicides. In summary, this PhD provided three case studies showcasing the value of data analytics across three different datasets that are related to text-based technologies and mental health: 1) showing the application of machine learning on a large text-based happiness dataset; 2) discovering the attitudes of professionals towards mental wellbeing chatbots (a modern text-based interactive technology); and 3) using data analytics to gain insight into the responsible reporting of suicide and mental health in text-based media articles.

Date of Award	Mar 2025
Original language	English
Supervisor	Edel Ennis (Supervisor), Raymond Bond (Supervisor) & Maurice Mulvenna (Supervisor)