Understanding loneliness through data-driven approaches: from practitioner insights to algorithmic detection
Publication Date
July 27, 2026
Creator
Milligan, Gregor
University of Nottingham
Abstract
Loneliness can be both a cause and symptom of broader mental health concerns, in addition to having established negative impacts on physical health, employability and educational outcomes. This thesis examined the driving factors of loneliness through the lived experiences of lonely individuals and mental health practitioners. To achieve this understanding, the thesis employed a mixed-methods approach, combining thematic analysis of practitioner interviews, natural language processing (NLP) analysis of therapeutic transcripts, quantitative analysis of survey data and classification of forum posts through machine learning (ML) techniques.
The first empirical chapter consisted of semi-structured interviews with digital mental health practitioners, which revealed the hidden prevalence of loneliness concerns in digital therapeutic contexts. Despite loneliness affecting the majority of clients, the thematic analysis showed that practitioners consistently encounter almost exclusively indirect disclosures of loneliness due to the stigmatised nature of the concern. Practitioners identified subtle patterns of social disconnection rather than explicit statements of loneliness, highlighting a core challenge in clinical practice.
The second empirical chapter analysed 254 therapeutic transcripts using NLP methods and established a range of motivations that service users have for engaging in digital mental health interventions. Workshops with clinicians and service users led to the development of a validated 12-item outcome measure. Findings revealed the same indirect expression patterns that practitioners noted in the first empirical chapter, confirming both the latent nature of loneliness disclosures and their prevalence across digital therapeutic settings.
The third empirical chapter analysed survey data across London, United Kingdom (n = 2886), exploring the relationship between loneliness and stigmatised deprivation indicators, including food and energy insecurity. Hierarchical ML modelling revealed that stigmatised deprivation significantly impacts loneliness beyond conventional socioeconomic measures, demonstrating that specific hardships create additional psychological burden independent of general income effects.
The final empirical chapter examined NLP detection methods for online loneliness disclosures. Traditional and contemporary NLP techniques were compared on labelled social media posts, while a systematic evaluation of data collection practices exposed critical methodological issues across the research domain. The results showed that high-performing classification models learn domain-specific linguistic patterns rather than genuine psychological indicators, revealing fundamental validity problems in current computational mental health research.
In conclusion, this thesis addresses critical gaps by revealing that loneliness is pervasive yet latent in therapeutic settings and that specific forms of deprivation create independent psychological impacts. Furthermore, computational detection methods can be susceptible to systematic data-confounding issues. These contributions have significant implications for both clinical practice and computational mental health research, while also establishing new frameworks for identifying loneliness and highlighting the need for rigorous methodological validation in the field.
Item Type
ethesis
Thesis Type
PhD
Supervisors
Subjects (LC)
Associated Schools / Departments
UoN Repository URI
Except where otherwise noted, this item's license is described as
File(s)![Thumbnail Image]()
Name
Gregor_Milligan_Thesis_Corrections_11_02_2026.pdf
Size
6.87 MB
Format
Adobe PDF
Checksum (MD5)
a98b98cb877d5853ac9047aac1f5aa69