Using AI to understand audiences

Jonathan Knott, 14.02.2017
Can machine learning help museums keep track of visitors?
In order to engage with their audiences, museums need to understand who they are and what makes them tick. But this is often easier said than done.

Even gathering basic information such as visitor numbers can present practical difficulties, especially for museums that have few resources, don’t charge for entry, or have a large amount of outdoor space.

Last year, researchers from the innovation charity Nesta carried out research exploring whether machine learning could help museums estimate their visitor numbers using social media data.

The project studied around 1,300 Accredited museums in the UK, a fifth of which had data on annual visitor numbers. It also referenced data from Arts Council England (ACE) on these institutions’ governance, Accreditation status, and location, and the government’s Multiple Deprivation Index, which provides information on the local economy and environment. These two sources were used alongside data from FourSquare, a social network that allows people to virtually ‘check in’ when they visit a venue.

By analysing the relationships between the different kinds of data, the project found that FourSquare check-in information could be used to help predict the approximate size of annual visitor numbers for museums that did not have this information. The model used did not provide actual figures in its visitor number predictions, but instead placed the museums into one of three broad visitor number categories (small, medium or large).

John Davies and Antonio Lima, the Nesta research fellows who carried out the project, found that their model could correctly predict the correct size category for more than half (57%) of the museums in a test sample.

In a blog describing the project, they concluded: “In general we find that the FourSquare check-ins data lets us predict the approximate number of museum visits much better than we otherwise could.”

“These machine-learning models are learning the pattern that exists between FourSquare check-ins, museum characteristics and visitor numbers. That means that when we have a certain pattern of the first two, they can predict what the visitor numbers should be,” explains Davies.

While Davies and Lima recognise that the model they used was relatively simple, they said that there was great potential for museums to use new digital data sources in this way.

Andrew Larking, the creative director at digital agency Deeson, who has previously worked at the Natural History Museum and the Science Museum, believes that computers offer the best way for museums to understand their visitors.

“If you look long enough, every action can be measured and analysed, and therefore predicted,” says Deeson.

“AI affords ways to quickly pull insights and understanding out of data, and to predict future events.”

Brendan Ciecko, the chief executive of Cuseum, a US company that provides the technology for cultural institutions to create mobile visitor apps, agrees that machine learning could enable museums to improve their performance in areas including marketing, fundraising and operations.

“A model could analyse a number of different variables like past attendance, weather, traffic patterns, day of the week, and other events in the area in order to make an educated prediction around what the attendance might be for a specific day or event,” he says.

Ciecko points out that customer relationship management (CRM) software like Salesforce already uses pattern recognition to help sales teams answer questions like the most effective times to contact leads. “A lot of that same notion can be applied to membership and fundraising,” he says.
 
“When you are dealing with datasets of tens of thousands of people, it just is not feasible for a human to pick up on things that a computer would be able to,” he adds.

 “A computer can note every single touch point, every single characteristic of a specific patron, what events they’re going to, what their giving level is, their age, and compare this to historic patterns to make predictions.”

Ciecko stresses that it is “extremely early days” regarding museums’ use of artificial intelligence (AI) techniques. But he says museums that want to establish momentum in this area should try to identify where they can quickly gain tangible results.

“Not-for-profits and cultural institutions only have so many resources – and in the hierarchy of needs there are things that come far before anything that appears like a shiny new object,” he says.

“There needs to be a little bit more social proof of institutions reaping the benefits of AI before we start to see everybody putting some sort of spin on it.”

When organisations can demonstrate they are getting financial value from AI, the sceptics will be encouraged to get on board, says Ciecko. 

“If you’re able to build a financial case that AI can drive new revenue and increase the sustainability of an institution, then it’s much easier to justify.”

Comments