Recently, a large number of knowledge bases have been made available, which store factual information in form of relationships between entities by means of graphs, RDF triples or logical representations. This data is inherently relational and stored as “knowledge graphs” made up of millions of nodes and billions of edges, or as billions of facts in form of RDF triples.
The interest in performing prediction and learning tasks on this data led to the development of scalable statistical relational learning techniques. These tasks arise in many settings – characterized by complex and uncertain relationships among entities – such as semantic web, information extraction, social networks, diagnosis, network communication, computational biology, etc.
The tutorial will provide an overview of representation options for large-scale graph-structured or relational datasets, then it will describe the wide spectrum of machine learning techniques that were devised for reasoning and learning over this data. Finally, it will present the main application domains, from link prediction to entity resolution and ontology learning.
Acknowledgments. This work was supported by Gruppo Nazionale per il Calcolo Scientifico (GNCS-INdAM).