Daystream annotated text corpus of traffic messages

The Daystream Corpus is a dataset of 3541 traffic messages in which proper names (e.g. roads, lines, stops), their reference IDs (e.g. DHID, DLID, OSM-IDs), as well as relations (e.g. traffic jam, accident, rail replacement traffic) are annotated manually. The data set can be used as a training or test body for information extraction tasks such as self-name recognition, entity linking and relations extraction. Dataset statistics: <TABLE> <TR> <TH></TH> <th style=“text-align:left“>Twitter</th>” <TH STYLE=“TEXT-ALIGN:LEFT“>RSS</TH>” <th style=“text-align:left“>Total</th>” </TR> <TR> <td>docs</td> <TD> 2825</TD> <TD> 716</TD> <TD> 3541</TD> </TR> <TR> <td>tokens</td> <TD> 69188</TD> <TD> 34630</TD> <TD> 103818</TD> </TR> <TR> <td>entities</td> <TD> 15280</TD> <TD> 8112</TD> <TD> 23392</TD> </TR> <TR> <td>relations</td> <TD> 365</TD> <TD> 427</TD> <TD> 792</TD> </TR> <TR> <td>docs with annotated relations</td> <TD> 305</TD> <TD> 338</TD> <TD> 643</TD> </TR> <TR> <td>linked entities (org|loc)</td> <TD> 5138</TD> <TD> 3331</TD> <TD> 8469</TD> </TR> <TR> <td>NIL entities</td> <TD> 4764</TD> <TD> 1698</TD> <TD> 6462</TD> </TR> </TABLE> The Daystream body is released under the CC-BY 4.0 license. If you use this data, you should quote the following publication: A German Corpus for Fine-Grained Named Entity Recognition and Relation Extraction of Traffic and Industry Events. Martin Schiersch, Veselina Mironova, Maximilian Schmitt, Philippe Thomas, Aleksandra Gabryszak, Leonhard Hennig. Proceedings of LREC, 2018. Further information and details: https://github.com/DFKI-NLP/daystream-corpus/