A seven layer DNN approach for social-media multilevel image-based text classification
Veena Tripathi, Navin Mani Upadhyay, Saurabh Sharma · 2024
In recent years, social media platforms have played a significant role in disaster management. To extract valuable insights and understand the meaning behind social media text content, using conventional machine learning methods, text mining solutions have been created. Through these techniques, messages are intended to be categorized into several themes, such counsel and caution. Nevertheless, these methods frequently focus on a single event and have trouble generalizing to cross events classifications. The effectiveness of traditional models trained on historical data for classifying social media messages from future events remains unclear. Our research focuses on the efficacy of a CNN model in classifying Twitter topics across diverse events. Three geotagged datasets from twitter gathered during the COVID-19 pandemic&s;s management are the subject of this study. Two conventional machine learning techniques (LR and SVM) are contrasted with the CNN model&s;s performance. Our experiments conclusively show that the CNN model significantly surpasses the performance of both SVM and LR models in terms of accuracy, regardless of whether the evaluation is based on individual events or across different events. This implies that the CNN model may efficiently categorize messages for impending events and pre-train on Twitter data from past occurrences, improving situational awareness. In conclusion, this study demonstrates the CNN model&s;s capability compared to traditional machine learning techniques for disaster management.