Semi-automatic ontology extension in the maritime domain

One of the tasks of a maritime safety and security (MSS) system is to map incoming observations in the form of sensor data onto existing maritime domain knowledge. This domain knowledge is modeled in an ontology. The sensor data contains information on ship trajectories, labeled with ship types from this ontology. These ship types are broad and within one type there can be several distinctive behavior patterns. As a consequence we cannot make a good mapping from these trajectories to the ship types. To make this possible we should change the ontology by adding relevant subtypes. This paper presents a semi-automatic method to extend the ontology of ship types on the basis of trajectory data. The first part involves the use of hidden Markov models to model the data of each ship within one ship type and the clustering of these models. The clusters are input to the second part where we use internet querying and natural language processing based ontology extension techniques to extend the maritime domain ontology. We present the promising results of a preliminary experiment that shows an interesting possibility in terms of semi-automatic ontology extension, which would enable an optimal coverage of a given domain: not providing too many concepts, and not leaving essential ones out.