Twitter is the well-known mass social network for communication among the individuals, groups, community, or forums on special occasions in the form of tweets. Sentiment analysis of Tweets is one of the essential requirements on the societal disaster events, among these tweets which have significant consequence on societal action measures. Our proposed study exhibits the tweet sentiment analysis on disaster messages using Naïve Bayes model and the two popular transformer models, namely Bidirectional Encoder Representations from Transformers (BERT), a Robustly Optimized BERT Pretraining Approach (RoBERTa). The system performance results of these models are evaluated with the metrics of precision, F1-score, and recall, and the model performance outcomes are 70%, 89%, and 90%, respectively, for Naïve Bayes multimodal classifier, BERT and RoBERTa algorithms. The results conclude transformer models are better performing in terms of tweet sentiment analysis, due to its dynamic contextual embeddings and automated feature learning. Particularly, RoBERTa outperforms accurate functioning on public, real-time tweet opinion analysis on the Kaggle disaster tweet dataset rather than BERT, because of the enhanced pretraining phase. The study insights could be used further for the disaster management in the perspective of humanitarian action progress.