LORE: a model for the detection of fine-grained locative references in tweets

Abstract

Extracting geospatially rich knowledge from tweets is of utmost importance for location-based systems in emergency services to raise situational awareness about a given crisis-related incident, such as earthquakes, floods, car accidents, terrorist attacks, shooting attacks, etc. The problem is that the majority of tweets are not geotagged, so we need to resort to the messages in the search of geospatial evidence. In this context, we present LORE, a location-detection system for tweets that leverages the geographic database GeoNames together with linguistic knowledge through NLP techniques. One of the main contributions of this model is to capture fine-grained complex locative references, ranging from geopolitical entities and natural geographic references to points of interest and traffic ways. LORE outperforms state-of-the-art open-source location-extraction systems (i.e. Stanford NER, spaCy, NLTK and OpenNLP), achieving an unprecedented trade-off between precision and recall. Therefore, our model provides not only a quantitative advantage over other well-known systems in terms of performance but also a qualitative advantage in terms of the diversity and semantic granularity of the locative references extracted from the tweets.