Explaining Commonalities of Clusters of RDF Resources in Natural Language

EasyChair Preprint 13143

Abstract

We introduce a system that provides explanations in Natural Language for individual clusters of RDF resources, where clusters are obtained using an external clustering tool. Our system is based on the theory of (Least) Common Subsumers (CS) in RDF. We propose an optimized algorithm for computing a CS, which allows us to compute the CS for up to 80 RDF resources (each with its own RDF-graph of linked data). We then generate a Natural Language sentence to describe each cluster. A unique aspect of our explanations is the use of relative sentences, including nested ones, to represent blank nodes in an RDF-path. We demonstrate the usefulness of our tool by describing the resulting clusters of a real, publicly available, dataset on Public Procurements.

Keyphrases: Explainable Artificial Intelligence (XAI), Least Common Subsumer (LCS), Natural Language Generation (NLG), Resource Description Framework (RDF)

@booklet{EasyChair:13143,
  author    = {Simona Colucci and Francesco M Donini and Eugenio Di Sciascio},
  title     = {Explaining Commonalities of Clusters of RDF Resources in Natural Language},
  howpublished = {EasyChair Preprint 13143},
  year      = {EasyChair, 2024}}