Qbiz UK conducts a Natural Language Processing (NLP) project at the London Fire Brigade

Dan Singh was embedded in the LFB’s analytics team for 2 months and worked alongside LFB analysts to explore the work.

Head of Data Analytics at LFB, Apollo Gerolymbos comments: “It was an absolute pleasure having Dan work with us. LFB collect mountains of data about all their activities and expanding into data science to make sure we aren’t missing any insights is a natural progression for a world class fire service. We hope to be able to share this approach and drive the whole sector forward.”

The first element involved expanding existing topic modelling capabilities to be applicable to new and existing datasets. Topic modelling is useful in identifying categories from unlabelled data.

For example given a dataset containing the text from Wikipedia articles on the top grossing films of all time, one could categorise their genres and highlight the words associated with each genre.

The topic modelling approach used was Latent Dirichlet Allocation (LDA), and the code was created in a format that will enable LFB to run LDA on any chosen set of articles. These could be fire investigation reports, reports about health and safety events or even consultation responses.

The second part involved building a predictive model to predict categorisation fields from article text. A logistic regression classifier was created with a predictive accuracy of over 70%.