A fire and rescue service in a city of more than 8 million people.
Between 1987 and 2016 the city’s fire and rescue service saw the number of fires fall, reducing the total number of annual fatalities dramatically.
At each incident staff record information for over 70 categorical fields in a database, while more serious incidents result in a report written by a fire investigator. Since 2000 the service has generated over 37,000 of these written reports. Such a large quantity of free text data has the potential to offer important insights into fire types, but reading through them all would take a human almost 2,000 hours. Yet thanks to recent improvements in computational understanding of language, extracting insight from a large collection of text data is now possible.
Working with the fire and rescue service, ASI used latent Dirichlet allocation to analyse all the text gathered from the 37,000 reports. The algorithm identifies topics in the documents, and in this way it was able to extract some coherent meaning from the large text data set.
The topic analysis distinguished between twenty different categories of fire, such as fires in heating systems and boilers, fires caused by faults in electrical appliances, or fires relating to commercial kitchens.
The project was able to confirm a hypothesis of the fire service that fires relating to the accumulation of grease and oil in ducting systems had been increasing over time. These fires were plotted on a map, identifying higher frequencies in Soho and on Edgware Road. This allowed the service to optimise their fire inspections and target premises that represented the highest risk.