Computer Science

Colloquium Details

Probabilistic-Logical Languages and the Web of Data

Abstract

An increasing number of datasets are published according to the principles of Linked Data. These principles formalize the integration of open data repositories using semantic links. The idea is that a large interlinked dataset is more than the sum of its parts. In most situations, however, finding meaningful links between heterogeneous datasets remains a difficult challenge. A data integration framework that combines logical and probabilistic reasoning has shown promising results. The approach leverages both numerical similarity scores and the logical structure of the datasets. In this talk, I will present the principles and applications of the framework. I will also discuss situations in which we do not have sufficient structure and introduce methods for learning logical schemas using statistical as well as crowdsourcing methods.

Biography

Mathias Niepert is a post-doc at the University of Mannheim, Germany, working in the Knowledge Representation and Management Research Group. He received his PhD from Indiana University Bloomington in 2009. His research areas include probabilistic graphical models, statistical relational learning, digital libraries and, more broadly, the (social) semantic web. Recently, he has been working on developing probabilistic logics for data integration, ontology learning, and activity recognition. Mathias received best paper awards at the international conferences UAI and ESWC. He is also co-founder of InPhO (Indiana Philosophy Ontology), a digital humanities project integrating statistical text processing algorithms, logic programming, and user feedback to build and maintain a computational ontology.