Knowledge Engineering and DIscovery (KEDI)

The Knowledge Engineering and DIscovery research group makes sense of data. We extract simple statements from unstructured texts or complex linked data sets, integrate them into consolidated perspectives and derive new facts and their explanations from that. This way we empower people with disruptive knowledge discovery methods, focusing especially on the domains of life sciences and digital humanities.

The research of the KEDI group focuses on:

Data Integration
Graph-based Knowledge Representation
Graph Mining and Statistical Relational Learning

“Our Knowledge Engineering activities intend to help data consumers and publishers to manipulate the data by providing innovative ways to search for relevant datasets/vocabularies and by providing new insights into their own data/vocabulary use.”

Knowledge Engineering

Knowledge Graphs – Graphs are useful for modeling objects, their relationships to each other, and the conceptual structures wherein which they lie. In our research, we explore the possibility to use both graph and geometric interpretations of the knowledge graph to support new type of analysis.

Ontology Design – The adoption of Semantic Web technologies (RDF, RDFS, OWL) has lead to the emergence of large datasets using richer representation models capturing the data semantics. The contribution of our research is to promote and facilitate the reuse of well documented ontologies or vocabularies in the Linked Data ecosystem.

Related technology: Linked Open Vocabularies

Linked Data Publication – Data providers have so far published hundreds of linked datasets and billions of triples on the Web, thus contributing to the birth of the so-called “Web of Data”. Nevertheless, Linked Data publication still faces problems that hinder its adoption. Our research focuses on monitoring and providing data publishers with empirical evidence of use and accessibility of their data.

Related technology: SPARQLES, LOD4All

“With knowledege graphs, scientists have now the means to aggregate semantic information from various sources. The challenge is not only to extract meaningful information from this data, but to gain knowledge, to discover previously unknown insight, look for patterns, and to make sense of the data.”

Knowledge Discovery

Graph Summarisation and Serendipitous Browsing – With more than millions nodes and edges, it is, for a human, almost impossible to understand the information encoded in large graphs. We are investigating graph summarisation methods to help users extract and understand the underlying information. Related technology: Semantic Text Exploration, Linkspire

Statistical Relational Learning – While a large majority of Machine Learning techniques focus on propositional data which is assumed to be identically and independently distributed, Many real world datasets are fundamentally based on complex relational structure where the data distribution is neither identical nor independent. An emerging research area, Statistical Relational Learning (SRL), attempts to learn in relational domain.

Distributional Data Semantics – Distributional data semantics is a bleeding edge field of research that aims at providing insights into the meaning of large and heterogeneous data sets in a bottom-up manner, using the implicit patterns emerging from the data themselves. The main contribution of our research consists of combining the classical distributional principles with graph analysis methods and symbolic reasoning. Related technology: CORAAL, SKIMMR

Tools & demos

Publications View all »

Latest News View all »

Translating Embeddings (TransE) – Knowledge Graph Mining

TransE is a textbook case link prediction model in multi-relational data (knowledge graphs). Learn more about Link Prediction models in this post presenting in a simple way Translating Embeddings (TransE), a method for the prediction of missing relationships in knowledge graphs: