A Computational Linguistic Exploration to Characterise Risk-Taking Behaviours in Bipolar

Harvey, Daisy and Jones, Steven and Rayson, Paul and Lobban, Fiona and Palmier-Claus, Jasper (2025) A Computational Linguistic Exploration to Characterise Risk-Taking Behaviours in Bipolar. PhD thesis, Lancaster University.

Text (A Computational Linguistic Exploration to Characterise Risk-Taking Behaviours in Bipolar)
2025harveyphd.pdf - Published Version
Restricted to Repository staff only until 12 March 2026.
Available under License Creative Commons Attribution-NonCommercial.
Download (4MB)

Abstract

Background Bipolar disorder (bipolar) is a severe mental illness affecting approximately 2% of the global population. It is associated with challenges in emotion regulation and cognitive processing, often contributing to risk-taking behaviours such as hypersexuality, excessive spending, and self-harm. Despite clinical and academic recognition of these behaviours, there is limited research on real-world experiences of risk-taking in bipolar. Existing studies predominantly rely on self-report measures or laboratory tasks, which fail to capture the full reality and impacts of these behaviours. This gap in understanding hinders the development of symptom-specific support and interventions. Objectives This thesis aims to characterise the specific types and frequency of risk-taking behaviours in individuals with bipolar disorder. By employing computational linguistic methods, it seeks to provide novel insights into these behaviours and inform healthcare providers on how to better address risk-taking, an under-recognised aspect of bipolar in clinical settings. Methods The research consisted of three original research studies: 1. Semi-structured interviews were conducted with 18 individuals with bipolar and five healthcare professionals to identify self-reported or observed risk-taking behaviours. Content analysis, combined with corpus linguistic techniques produced a classification system encompassing 39 behaviours across six domains. 2. Natural Language Processing (NLP) methods were applied to produce a corpus of Reddit posts, curated through web scraping, to further explore one of the identified risk-taking behaviours from Study 1: hypersexuality. This involved topic modelling, psychological domain analysis, and keyword, concordance, and collocation analyses. 3. A focused investigation into hypersexuality in bipolar was undertaken, informed by interview findings and recent literature, using computational methods to examine its presentation and the discussion of hypersexuality on Reddit by people who self-report a diagnosis of bipolar. Results The interviews and subsequent analyses identified 39 risk-taking behaviours across six domains, providing a comprehensive classification system. NLP analysis of Reddit posts revealed a significant increase in hypersexuality-related discussions over the past decade. Qualitative findings highlighted a lack of clinical awareness and support for hypersexuality, suggesting an unmet need for understanding and addressing this symptom. Conclusion This thesis underscores the importance of normalising discussions around risk-taking behaviours, particularly hypersexuality, in the management of bipolar. By integrating computational linguistic methods with qualitative insights, it offers a novel framework for understanding risk-taking behaviours. The findings have implications for improving clinical assessment, intervention, and the overall management of risk-taking in bipolar.