Gerhard Wohlgenannt 

Date/Time: 16.09.2019, 16:00 

Location: D2.2.094

Abstract

In this talk we briefly discuss two independent topics in a practical way. Firstly, while document classification is a well-known problem in NLP and IR, extreme multi-label document classification, ie. the multi-label classification of a large number of documents with a large number of labels and a skewed label distribution is very hard and largely unsolved. We will discuss the application of a modern deep learning toolkit (fastAI) to this problem and some tweaks to raise classification performance. Secondly, the talk will introduce visual question answering over Linked Data with an ontology visualization tool (Ontodia), and its potential combination with existing QALD (question answering over linked data) tools. 

Bio 

Gerhard Wohlgenannt is holding a fellowship position as research professor at ITMO university in St. Petersburg, Russia, where he is working at the "Faculty of Software Engineering and Computer Systems".

Before, Gerhard Wohlgenannt was an assistant professor at the Institute of Information Business of the Vienna University of Economics and Business, Austria.

He completed his PhD thesis in the field of ontology learning on a method for learning ontology relations by combining corpus-based techniques and reasoning on data from Semantic Web sources in 2010. In July 2016, he received his habilitation degree (venia docendi) in Business Informatics.

His research interests are knowledge extraction from text, natural language processing, ontology learning, and crowdsourcing.