Skip to Main Content
In this paper we examine effective representations of knowledge for the purposes of management of engineering and technology. Specifically, given the immense volume of data available about scientific outputs, it is highly necessary to condense or abstract this information for management use. This paper considers the utility of such representations in the management of technology. We ask further whether a given representation accurately depicts the knowledge contained in the science and technology database. We argue that, in this regard, generative models are superior because they provide explicit hypotheses about the structuring of the data. The second is whether the representation is interpretable by management, and therefore directly actionable. We argue that the number of model parameters is an indirect measure of the degree of difficulty of using and interpreting the selected representation. Combining the two metrics suggests the use of Akaike's Information Criteria, a metric used for model selection purposes. The AIC is used to evaluate existing model representations used in tech mining, both positional and relational. After surveying the results, we recommend the use of a mixed representation. These more complex models appear to offer a more useful representation of science and technology datasets. Furthermore there are multiple promising but previously unexplored representations of the data. The ramifications of further exploration within this range of possible new models is discussed.