Bootstrapping patterns for the detection of mobility related events

Britta Grusdt, Jan Nehring, Philippe Thomas

In: 14th Conference on Natural Language Processing. Konferenz zur Verarbeitung natürlicher Sprache (KONVENS-2018) September 19-21 Vienna Austria Verlag der Österreichischen Akademie der Wissenschaften 2018.


This work presents a method to extract traffic events from German texts. We present a rule based system, where patterns are automatically extracted and ranked using a bootstrapping approach. These patterns are subsequently evaluated and annotated by human annotators. The resulting pattern are evaluated on three different text sources (Tweets, traffic RSS feeds, and news articles) with different language styles. Through the use of three data sets we cannot only evaluate the usefulness of the approach in a single domain but also evaluate the domain portability of the proposed approach. We further perform an error analysis to identify problems of the current system.


KONVENS_2018_paper_20.pdf (pdf, 707 KB)

German Research Center for Artificial Intelligence
Deutsches Forschungszentrum für Künstliche Intelligenz