A Mobile Intelligent Dialogue Agent for gas classification in crisis situations

This thesis presents a Mobile Intelligent Dialogue Agent that can be used in the crisis management project DIADEM. The Agent is used in crisis situations where a chemical accident near a populated area has occurred and a gas is released. The Agent is implemented as a mobile application and can interview people on their mobile phone to gather information on what they smell. Research is done on olfactory perception to see how people describe odors and a gas database is created where gases are described in more detail. A Bayesian Belief Network then is implemented to classify what people smell using the answers from the people. The Agent uses a simple interview style where only ”yes/no” questions are asked. This way of interviewing people to gather information is faster then the current method, which is over the phone. Next to this, the Agent uses the Bayesian Network to select the most probable questions given the probabilities of the nodes in the network. In this way, the interview can adapt to the current crisis situation and the interview will be as short as possible. This has resulted in a first implementation of the Mobile Intelligent Dialogue Agent for the Android operating system. The Agent is able to select the most probable questions first and in theory can also adapt to the crisis situation.