Social networking sites have become a part of our daily lives. Due to the development of technology and usage of such networking sites, people come to know of the news alerts related to natural calamities very quickly. But the authenticity of such news needs to be taken care of. Deploying a physical sensor system in a residential area for earthquake detection is a tough and expensive solution. Hence, social network data can be used by humanitarian organisations to find help, and save lives. The primary goal of our work focus on near real-time earthquake detection.
Twitter has become a potential data source to explore useful information mentioned by users. For instance, Twitter tweets related to earthquake can be used to detect temporal occurrence as well as location information by a humanitarian organisation. The proposed system will use deep learning techniques such as RNN/LSTM to check the validity of Twitter tweets and real-time detection of the given event. In this work, machine learning models trained from tweets related to the earthquake in the past labelled by crowdsourcing plays a role as the classifier to predict the validity of the tweets. The system will use Twitters API to listen to particular keyword like earthquake and take these tweets as inputs. The tweets are converted to word embeddings using BERT before giving them to the model. The proposed system can detect the earthquake after it happens in the level of tolerance and ensure earlier warning to the public than any websites.
