Towards automated mental health first aid provision using Natural Language Processing

Owen, David ORCID: https://orcid.org/0000-0002-4028-0591 2025. Towards automated mental health first aid provision using Natural Language Processing. PhD Thesis, Cardiff University. Item availability restricted.

	PDF (David Owen, PhD Thesis) - Accepted Post-Print Version Restricted to Repository staff only until 12 March 2027 due to copyright restrictions. Available under License Creative Commons Attribution Non-commercial No Derivatives. Download (1MB)
	PDF (Cardiff University Electronic Publication Form) - Supplemental Material Restricted to Repository staff only Download (598kB)

Abstract

Mental illness is prevalent around the world. Depression and anxiety are particularly common and affect hundreds of millions of people. These diseases can cause hardship, leading to lower productivity and an increased risk of suicide. Early healthcare intervention is crucial to prevent the worsening of mental illness and to promote better long-term outcomes. However, due to the stigma associated with mental illness, people who might need to seek help from healthcare services are often reluctant to do so. Instead, they may turn to social media platforms such as Twitter or web-based forums such as Reddit to communicate their experiences. This study hypothesises that Artificial Intelligence (AI) can be used to identify users of these online communities who are at risk of developing mental illness so that they can be referred to appropriate help. Techniques from Natural Language Processing (NLP), a subfield of AI, were used to find linguistic features in text written by users in these online communities that may be indicative of depression and anxiety. NLP-based approaches demonstrated effectiveness in distinguishing users with depression or anxiety from those without. The results also suggested that the final 12 weeks (about 3 months) of posts before a depressed user’s eventual diagnosis by a clinician are the most indicative of their illness. Using this finding, it may be possible to direct people with depression to a clinician or other appropriate help much sooner than otherwise. To improve approaches to detecting depression and anxiety in users, it is likely that multimodal methods should be pursued. Such methods would consider, for example, images posted by users in addition to the text they write. The Mental Health Triage Scale (MHTS) could be used for decision making therein to direct users to the help they need.

Item Type:	Thesis (PhD)
Date Type:	Completion
Status:	Unpublished
Schools:	Schools > Computer Science & Informatics
Subjects:	Q Science > QA Mathematics > QA75 Electronic computers. Computer science
Date of First Compliant Deposit:	12 March 2026
Date of Acceptance:	10 March 2026
Last Modified:	12 Mar 2026 15:31
URI:	https://orca.cardiff.ac.uk/id/eprint/185709

Actions (repository staff only)

Edit Item