Functional Data Analysis and Knowledge-Based Systems

In the present study, the challenge is whether a distant reading of the history of a discipline can be achieved by analysing the temporal evolution of keywords retrieved from papers in the discipline’s mainstream journals. This calls for a so-called knowledge-based system (KBS), i.e., a computer-based system that supports human learning not only by acquiring and manipulating large volumes of data and information, but also by integrating knowledge from different sources. In this chapter, we introduce a KBS that, starting from a large database of texts retrieved from scientific articles published over a lengthy period by a selection of the discipline’s premier journals, leads to the construction of a well-founded corpus of scientific literature and from this to a possible outline of the discipline’s history. Our work is based on the idea that the temporal course of a word occurrence is a proxy of the word’s life cycle. We then adopt a functional data analysis (FDA) approach under which we first reconstruct words' life cycles. Second, by clustering words with similar life cycles, we detect any prototypical or exemplary temporal patterns representing the latent dynamics of word micro-histories. The major dynamics uncovered at this stage are then submitted to subject matter experts for interpretation and guidance in decision-making, thus making it possible to trace a history of the discipline. Moreover, we propose several kinds of data normalisation which involve different concepts of life cycle similarity and hence a different reading of the history of the discipline under examination.