Analysis and Cluster Based Modelling and Recognition of Context in a Mobile Environment

Abstract

Abstract. Our intuitive understanding of context has been formulated by Dey, as the information describing the situation of an entity. In the case of simple information sources, determining the relevant information describing the situation or context of an entity is straightforward. This is not the case when several information sources can be used, essentially to generate high level contexts. We set out to show that extracting the relevant context information from the combination of multiple information sources can be achieved by considering contexts as clusters in the data. We demonstrate this by analyzing a set of real measured data. A cluster structure in the data is visualized and it is shown how different user contexts are associated with different clusters. The cluster structure of context data can be modelled using a dynamic mixture model which gives insight into which properties a clustering algorithm should have in order to be used in a context recognition application. 1