The growth of online health communities and socially generated health-related content has the potential to provide considerable value for patients and healthcare providers alike. For example, members of the public can acquire medical knowledge and interact with others online. However, the volume of information-and the consequent 'noise' associated with large data volumes-can create difficulties for users. In this paper, we present a data-driven approach to better understand these data from multiple stakeholder perspectives. We utilise three techniques-sentiment analysis, content analysis, and topic analysis-to analyse user-generated medical content related to Lyme disease. We use a supervised feature-based model to identify sentiments, content analysis to identify concepts that predominate, and latent Dirichlet allocation strategy as an unsupervised generative model to identify topics represented in the discourse. We validate that applying three different analytic methods highlights differing aspects of the information different stakeholders will be interested in based on the goals of different stakeholders, expert opinion, and comparison with patient information leaflets.
Keywords: Lyme disease; content analysis; machine learning; online health communities; sentiment analysis; topic analysis.