What Is Semantically Important to “Donald Trump”?

Abstract

In the recent years, there is a growing interest in combining explicitly defined formal semantics (in the forms of ontologies) with distributional semantics “learnt” from a vast amount of data. In this paper, we try to bridge the best of the two worlds by introducing a new metrics called the “Semantic Impact” together with a novel method to derive a numerical measurement that can summarise how strong an ontological entity/concept impinges on the domain of discourse. More specifically, by taking into consideration the semantic representation of a concept that appears in documents and its correlation with other concepts in the same document corpus, we measure the importance of a concept with respect to the knowledge domain at a semantic level. Here, the “semantic” importance of an ontology concept is two-fold. Firstly, the concept needs to be informative. Secondly, it should be well connected (strong correlation) with other concepts in the same domain. We evaluated the proposed method with 200 BBC News articles about Donald Trump (between February 2017 and September 2017). The preliminary result is promising: we demonstrated that semantic impact can be learnt: the top 3 most important concepts are Event, Date and Organisation and the least essential concepts are Substance, Duration and EventEducation. The crux of our future work is to extend the evaluation with larger datasets and more diverse domains.