Automatised analysis of emergency calls using Natural Language Processing
Emarin Andersson, Benjamin Eriksson, Sofia Holmberg, Hossein Hussain, Lovisa Jäberg, Erik Thorsell · Gothenburg University Publications Electronic Archive (Gothenburg University) · 2016
The operators at SOS Alarm receives thousands of calls each day at the different emergency medical communication centres, owned by SOS Alarm, all over Sweden. A subset of these calls contain room for improvement and the operators could learn to improve from these calls. The work of finding – and analysing – these calls is however too tedious to be done by a human. This thesis presents four automatised solutions to this issue. The human factor is removed and the job of finding and analysing the calls is done by a computer. It is shown that it is possible to partly automatise the analysis, but the methods used have different strengths and weaknesses. Word frequency analysis is proven adequate at key word lookup. Similarity comparisons of various aspects of the calls are proven good at classifying calls, but less good at answering specific questions. Comparing parse trees seems promising, but the technology needs more work before it is ready to be used. The solutions presented show that it could be possible to automatise the analysis of the calls given that the right questions are asked and the results from these are used appropriately.