Does how we tweet convey our aesthetic preferences and artistic interests? The Barbican were interested in how they could better understand their audiences using data from Twitter. In particular, we were interested to see how tricky it would be to model the diversity of interests and arts based preferences among their followers, i.e., can we identify those who are more culturally curious from the way they tweet? And can we do this as automatically as possible? In order to do this I applied techniques from natural language processing to analyse audience behaviours and interests.
The Barbican, as a multi arts centre, curates a programme of events and exhibition that spans theatre, music, dance and visual arts, and a diverse mix of genres. The Barbican is also an iconic London Landmark, and the way in which people talk about the Barbican includes simply letting follows know they are there, sharing pictures of the striking architecture, the tropical trees in the conservatory or their lunch time interlude by the waterside, to more programme focused discussions about what film, gig, exhibition or show they are checking out.
Before getting down to any analysis there was the necessary – and not to be underestimated – data cleaning and filtering. I now a lot more about the Barbican Road area of Jamaica (which seems to be the spot for a lot of big club nights), the oft talked about locus of wild nights out in Plymouth, the Barbican waterfront and a very popular malt beverage, called -you guessed it – Barbican.
Using a combination of supervised and unsupervised machine learning approaches I classified conversational topics from the Twitter data and located potential `boundary crossers’ among followers (those whose tweets covered a broad range of the automatically extracted topics), i.e. those partial a spot of concerto and a dash of drum ‘n’ bass.
This project was a collaboration with Chatterbox Labs and the Barbican, funded by CreativeWorks London.
I presented the work at Digital Intelligence 2014 in Nantes and received the Best Paper Award.