A German Corpus for Fine-Grained Named Entity Recognition and Relation Extraction of Traffic and Industry Events

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 DFKI GmbH Berlin, Germany {firstname.lastname}@dfki.de Abstract Monitoring mobility- and industry-relevant events is important in areas such as personal travel planning and supply chain management, but extracting events pertaining to specific companies, transit routes and locations from heterogeneous, high-volume text streams remains a significant challenge. This work describes a corpus of German-language documents which has been annotated with fine-grained geo-entities, such as streets, stops and routes, as well as standard named entity types. It has also been annotated with a set of 15 traffic- and industry-related n-ary relations and events, such as accidents, traffic jams, acquisitions, and strikes. The corpus consists of newswire texts, Twitter messages, and traffic reports from radio stations, police and railway companies. It allows for training and evaluating both named entity recognition algorithms that aim for fine-grained typing of geo-entities, as well as n-ary relation extraction systems. Keywords: Named Entity Recognition, Relation Extraction 1. Introduction Monitoring relevant news and events is of central impor- tance in many economic and personal decision processes, such as supply chain management (Chae, 2015), mar- ket research (Mostafa, 2013), and personal travel plan- ning (Schulz et al., 2013). Social media, news sites, and also more specialized information systems, such as online traffic and public transport information sources, provide valuable streams of text messages that can be used to improve decision making processes (Hennig et al., 2016). For example, a company’s sourcing depart- ment may wish to monitor world-wide news for dis- ruptive or risk-related events pertaining to their suppli- ers (e.g. natural disasters, strikes, liquidity risks), while a traveler wants to be informed about traffic events re- lated to her itinerary (e.g. delays, cancellations). To ful- fill such information needs, we need to extract events and relations from message streams that mention fine- grained entity types, such as companies, streets, or routes (Yaghoobzadeh and Sch¨utze, 2017; Shimaoka et al., 2017). For example, from the sentence “Berlin: Rail re- placement service between Schichauweg and Priesterweg on route S2”, we would like to extract a Rail Replace- ment Service event with the arguments location=“S2” of type location-route, and start-loc=“Schichauweg” respec- tively end-loc=“Priesterweg” with types location-stop. Detecting such relations in textual message streams raises a number of challenges. Social media streams, such as Twitter, are written in a very informal, not always gram- matically well-formed style (Osborne et al., 2014), which cannot easily be processed with standard linguistic tools. News sites provide well-formed texts, but their content is very heterogeneous and often hard to separate from non- relevant web page elements. Domain-specific informa- tion sources, like traffic reports, on the other hand, are topic-focused, but employ a wide variety of formats, from telegraph style texts to table entries. In addition, exist- ing corpora for German-language Named Entity Recogni- tion (Tjong Kim Sang and De Meulder, 2003; Benikova et al., 2014) are mostly limited to standard entity types, …

Content truncated to 500 words