Akademik Veri Yönetim Sistemi
INTERNATIONAL JOURNAL OF CREATIVE RESEARCH THOUGHTS, cilt.11, sa.5, ss.100-108, 2023 (Hakemli Dergi)
Earthquakes cause massive damage to people and structures. The capacity to quickly assess damage over a large area is critical for successful disaster response. In recent years, social networks have proven to be a great capability to increase situational awareness and identify affected areas. In this context, this paper presents a method for assessing damage levels in earthquake zones utilizing social media data and the Naive Bayes, Support Vector Machine (SVM), Random Forest, and BERT classification algorithms. In order to compare different machine learning models in this study, we utilized our post-earthquake damage classification dataset consisting of Turkish Tweets, which we prepared and labeled for the problem. Identifying damaged structures from Turkish tweets after the earthquake is the first step of the solution which we propose. For spatial analysis, we need to extract the address information of the damaged structures from the tweets. Therefore, in this paper, we use Named Entity Recognition for address extraction and fine-tune the pre-trained BERT model with our own compiled detailed address detection NER dataset. Finally, in order to make disaster response more successful and effective, we obtained the latitude and longitude of damaged structures in the earthquake zone by obtaining the geographical coordinates of the addresses using the geocoder API, the address information obtained with the help of the NER model. Thus, rescue teams can intervene more effectively and increase their success rates.