Twitter Sentiment Classification Using ESOA Based Feature Selection with MHAM-DMO Model

Opinion mining is the process of analyzing the content people create, such as product reviews or social media posts, to determine if the feelings expressed are positive, negative, or neutral. Twitter, one of the most popular social platforms for sharing opinions, provides a lot of data that can be used to understand public sentiment. In this project, we developed a system for classifying sentiments that begins with a detailed preprocessing step using natural language processing techniques. After the data has been processed, the tweets are represented using the traditional Term Frequency-Inverse Document Frequency (TF-IDF) model to highlight the most important text features.To make these features even more relevant, we introduced the Egret Swarm Optimization Algorithm (ESOA), a method for selecting important features inspired by how Great and Snowy Egrets hunt. ESOA uses three strategies—waiting patiently, actively searching, and making decisions based on differences—to find a good balance between exploring new areas and focusing on known ones. This creates a flexible framework that works well in different situations. For sentiment classification, we use a Multi-Head Attention Mechanism (MHAM) that can understand various meanings in user text. We fine-tuned the model’s settings using the Dwarf Mongoose Optimization (DMO) algorithm, along with a strategy that helps each part of the attention mechanism focus on different aspects of the text. Testing our approach on the Sentiment140 dataset shows it works very well, achieving almost 97% accuracy, which is better than other methods that usually reach between 92% and 95%.