Interpreting Arts Audiences and Cultural Preference Through Twitter Data

In this study we work with a multi-arts organisation to assess the appropriateness of natural language processing methods for the analysis of audience behaviours and interests via Twitter. We investigate supervised and unsupervised topic modelling methods to investigate whether they can: i) capture the nuanced differences between art genres/forms; ii) characterise users according to their cultural interests; and iii) help to surface the more culturally mobile members of the audience. We show promising results: supervised methods showed high accuracy (95%) in filtering data for relevance and pre-defined topics; unsupervised methods provided novel topic discovery, showing sophisticated genre groupings and discovering audience members with wider interests.