MobIE: A German Dataset for Named Entity Recognition, Entity Linking and Relation Extraction in the Mobility Domain

We present MobIE, a German-language dataset, which is human-annotated with 20\ncoarse- and fine-grained entity types and entity linking information for\ngeographically linkable entities. The dataset consists of 3,232 social media\ntexts and traffic reports with 91K tokens, and contains 20.5K annotated\nentities, 13.1K of which are linked to a knowledge base. A subset of the\ndataset is human-annotated with seven mobility-related, n-ary relation types,\nwhile the remaining documents are annotated using a weakly-supervised labeling\napproach implemented with the Snorkel framework. To the best of our knowledge,\nthis is the first German-language dataset that combines annotations for NER, EL\nand RE, and thus can be used for joint and multi-task learning of these\nfundamental information extraction tasks. We make MobIE public at\nhttps://github.com/dfki-nlp/mobie.\n