In recent years, sentiment classification in Twitter using deep learning approaches has gained popularity. Many researchers have focused on Twitter sentiment analysis and have made an assumption that all words within a tweet have the same polarity, often neglecting the polarity of individual words within the sentence. This paper proposes a novel approach to analyzing tweets, which consists of two main phases: feature selection and classification. In the first phase, the most appropriate features are selected through mutual information analysis. The second phase involves utilizing a Meta Heuristic algorithm to enhance the weights and biases of the multi-layer perceptron network. The study results demonstrate that the MLP network optimized by the Glow-worm Swarm optimization outperforms other existing methods.

Twitter Sentiment Analysis, Optimization, Feature Selection, Multilayer Perception.