• artificial intelligence
• network science
• machine learning
• reasoning
• semantic web
• knowledge graph
• instance matching
• ontology alignment
• knowledge graphs interlinking

Human beings have always tried to pass down knowledge to preserve it and let further generations exploit it and evolve. Being able to formalize knowledge allows us to exploit it to better understand actual reality, infer new facts, and predict future events.

Since the dawn of Artificial Intelligence, scientists have had an interest in methods for representing human knowledge and doing reasoning (creating the sub-field of Knowledge Representation & Reasoning) being these two characteristics strictly linked to human nature: to expressively describe what surrounds us and to infer new information from what's already known.
Knowledge Graphs (KGs), are one of the results of these scientific efforts: with these it's possible to represent human knowledge in the form of concepts and entities bridged by relations, enriched by the Ontologies that add a semantic layer that enables the possibility to do reasoning.

Nowadays, the use of KGs as background knowledge is widespread in Machine Learning (ML). Companies and users continuously create and share them: as a result, it is common that the same real-world entity can be described differently by different KGs.
Identifying identity relations is a challenging task, usually called Instance Matching (IM), that enables the interlinking of different and disjoint KGs with the aim of supporting the expansion of the knowledge available to Intelligent Systems and ML models. The IM can be seen as a special case of the Link Prediction (LP) task, in which the only edge type to be predicted between individuals of different KGs is the <owl:sameAs>.

In this thesis, I present a novel approach that involves reasoning, ML, and Network Science features that actually succeeds in overcoming the state of the art.