Flood Monitoring with Information Extraction Approach from Social Media Data

Abstract

Flood natural disasters that often occur in Jakarta have a bad impact on many sectors. Countermeasures, fast action, and monitoring need to be done to minimize the impact that occurs. Social Media is a technology platform that can provide flood-related data that can be used as primary data or complementary data for monitoring systems. This study focuses on using social media data to be used as flood monitoring data. The analysis used is an analysis with a natural language processing approach. The classification algorithm method used in this study is naive Bayes, random forest, support vector machine, logistic regression, and conditional random field. Location information extraction methods used are Standford NER and Geocoding. This research produces three models. The first model is the classification model used to classify relevant data with an f1- score evaluation value of 82.5%. The second model is the NER model which is used to extract location entities from sentences with an f1-score evaluation value of 73%. The last one is the locator of geocoding with a success rate of 75% for identifying roads. This research also produces a simple dashboard that can be used as a visualization tool.