The ever-increasing amount of language that is available online on social media sites and in Electronic Health Records (EHRs) provides a unique opportunity to understand more about mental health conditions. Applying Natural Language Processing (NLP) methods to this data allows us to study language empirically and provides an insight into typically hard to reach populations. This paper presents a PhD project which will investigate the lived experience of risk-taking for people living with bipolar disorder as presented in different contexts (on social media and in medical records), using a mixed methods approach encompassing qualitative and quantitative methods of analysis, and which aims to produce as the output of the research 1) a risk-taking lexicon for bipolar disorder 2) a corpus of risk-taking posts from bipolar subreddits and 3) linguistic analysis of the risk-taking behaviours extracted from free-text within anonymised EHRs. It is hoped that this research will shed light on how risk-taking behaviours manifest in bipolar disorder and whether there are differences in how these behaviours are described in online and offline settings.