Implementation of Named Entity Recognition (NER) for Spatio-Temporal Detection and Sentence Modeling of Natural Disasters Using Naive Bayes and C.45

Social media users share details regarding natural disasters, including the requirements of impacted communities and the locations of the disasters. The information can be used as one of there sources to map natural disaster events and the needs of disaster victims in Indonesia. However, information from social media has an informal structure and low credibility as an information provider. Unstructured social media data affects the identification of information related to the location, conditions, and needs of communities affected by natural disasters. Therefore, the formed tweets are used to create a text/sentence model to predict sentences related to natural disasters using the Naïve Bayes and C4.5 methods. Next, entity extraction is carried out to determine the location, date, and type of natural disaster using Statistic-based Named Entity Recognition (NER) using the Naïve Bayes method. The identification results of location, date, and type of disaster entities and texts/sentences related to natural disasters can help the National Disaster Management Agency (BNPB) to obtain early information related to natural disasters. The testing results for predicting sentences related to natural disasters using a dataset of 4,699 records showed that the Naïve Bayes method achieved an accuracy of 90.91%, while the C4.5 method achieved an accuracy of 93%. Additionally, with a dataset of 522 records, the Naïve Bayes method achieved an accuracy of 96.40%, and the C4.5 method achieved an accuracy of 86% The test results indicate that as the number of entity data testing increases, the accuracy value also increases for Naïve Bayes Method. Otherwise, for C4.5 the number of entity data testing increases make the accuracy value decrease. The NER accuracy values using Naïve Bayes for D-2 (12 records) are higher compared to D-1 (11 records), with the respective values being 81% and 91.60%.