Science is the conversion of information to knowledge by abstracting measured quantities either to validate existing models of reality or to construct new ones. In the new data intensive era, the hope is that computers will be able to perform this process using a combination of data mining and semantic technologies: for example, an unsupervised learning algorithm, such as a Kohonen map, could be applied to data to determine how many classes it is comprised of and this could then be automatically translated into a basic ontology for the data. The ontology could then be evolved to give a fuller description with the introduction of more data, genetic algorithms, or supervised learning techniques. In such approaches, the emphasis is usually placed on the class or concept representations in the formal model specification whereas contextualization is really the key to understanding data. In this paper, I will review current practises employed in describing relationships between data and show that a context-oriented approach to astronomical data gives a more powerful and flexible conceptualisation than is provided by an object-oriented one.