In the modern digital world, social media plays an essential part in increasing awareness and delivering information. Sentiment analysis is one of the most important challenges in NLP (Natural Language Processing) because of its complexity and impact on daily life. It relates to categorization based on the behavioural patterns in text by identifying behavioral characteristics. Disaster-related posts are collected from social media platforms such as Twitter, and analysis of emotions is used to determine their approach. In situations of disaster and emergencies, social media studies investigate whether first responders may use this information to improve crisis management and situational awareness. Social media posts can provide useful data for disaster management. They may be used to detect sentiments and identify the requirements of those affected by disasters. To enhance the effectiveness of text sentiment analysis, we restructured the behavioural model as an identical problem. Therefore, we introduce a MHA-BiLSTM (Multi-Head Attention with BiLSTM) model. Furthermore, classifying the text using situational or non-situational methods proves to be more efficient than traditional sentiment analysis. Bidirectional LSTM is used for initial feature extraction, while multi-head attention captures valuable information from different dimensions and representation subspaces. The classification mechanism scores the characteristic sources by comparing them with labeled sets. The outcomes of experiments indicate that the MHA-BiLSTM model outperforms many existing models on the Turkey and Syria Earthquake 2023 tweet sentiment analysis datasets.