Location extraction from Traffic Event-related Text

Abstract

One of the contents that will play an important role in the trip planning system is content related to geospatial. To ensure that the trip planning system is optimal and safe, information about traffic conditions, events, as well as their location is required. People often talk about traffic conditions on social media, such as traffic jams, detours, or accidents. These activities provide textual data related to traffic events. Location is regarded as essential information from traffic event-related texts to identify where the event took place. This study uses Indonesian short text for location extraction with named entity recognition (NER) technique. Data from twitter-based social media (lewatmana.com) are collected. Bidirectional Long Short-Term Memory - Conditional Random Field (BiLSTM - CRF) model and Indonesian POS tagger are used to develop the named entity recognition model for location extraction. Our current model shows promising results with 91.21% accuracy.