A temporal ontology guided clustering methodology with a case study on detection and tracking of artificial intelligence topics

Detection and tracking of topics from publicly available academic data can benefit the scientific community and other stakeholders throughout their investment and other decisions, by informing the decisions regarding the field of science, its evolution, and its dynamics. In this study, we introduce a novel temporal clustering method for topic detection, using document abstracts, keywords, and their corresponding textual representations. In this method, the temporal dimension is employed to parameterise the effect of older data on the clusters, while ontology guidance is utilised to guide their evolution. Ontology is used for both enhancing the representations, and decision-making for the evolutionary steps of split and merging of the clusters. We show the effectiveness of the representations of documents in a single time slice, before demonstrating the evolution of topics in a case study of AI-related publications. Finally, the resulting topic evolutionary map is evaluated after automatically labelling the clusters using ranked author keywords, facilitating the assessment of the topics and observing their evolution.