An Automatic Event Detection from Social Media Using an Improved Meta-Heuristic Model-Based Multi-Scale Deep Learning Network

Social media has become the primary and popular platform to share incidents that happen to every individual. Hence, the data generated by social media represents richer information by considering the ongoing events. Further, the timeliness associated with the data can provide valuable insights. Social media is one of the active communication models for connecting with each other, so event detection on social media is a widespread concern. Moreover, the deep learning models are introduced for detecting events based on temporal and spatial representations. Yet, focusing on the dynamic and large volume of data available in social media is impractical for filtering the events manually, and it is a time-consuming process. Further, the pre-trained models are less effective because of considering the mis-spelt words and enabling serious information loss. To alleviate these drawbacks in the conventional models, the research study employs a novel deep learning model for detecting the events in the social media effectively. This research proposed a multi-scale deep learning for event detection by analysing the social media data. The news agents or the decision maker can analyse the insight into the situation with the proposed event detection model. In the beginning, significant data are collected from publicly available datasets. Further, the gathered data is given to the text pre-processing stage for removing the redundancies to generate clean and quality data. Further, the pre-processed text is fed into the Optimal Weighted Multi-scale Transformer-based Deep Network (OWMT-DNet), in which the deep network is taken as Residual Bi-directional Long Short-Term Memory (Res-Bi-LSTM) for social event detection. In the proposed network, the weight optimisation is done by using the proposed algorithm named Opposition Piranha Foraging Algorithm (OPFA). Finally, the experimental validation is done for the recommended event detection model and compared with baseline approaches. The proposed approach attained an accuracy, precision, F1-score, and sensitivity of 97.27%, 97.29%, 90.10% and 97.25%, respectively, in event detection on social media. This performance improvement of the developed model gained valuable impacts and was applied in real-time monitoring, emergency response and analysing public sentiments. Moreover, the deep learning model can effectively analyse and identify the events in emergency response like natural disasters, accidents and public health emergencies.