Unsupervised clustering of context data and learning user requirements for a mobile device

Abstract

The K-SCM is an unsupervised learning algorithm, designed to cluster symbol string data in an on-line mariner. Unlike many other learning algorithms there are no time dependent gain factors. Context recognition based on the fusion of information sources is formulated as the clustering of symbol string data. Applied to real measured context data it is shown how the clusters can be associated with higher level contexts. This unsupervised learning approach is fundamentally different to the approach based, for example, on ontologies or supervised learning. Unsupervised learning requires no intervention from an outside expert. Using the example of menu adaptation in a mobile device, and a second learning stage, it is shown how user requirements in a given context can be associated with the learned contexts. This approach can be used to facilitate user interaction with the device. © Springer-Verlag Berlin Heidelberg 2005.