Semantic Feature Extraction-Based Twitter Sentiment Analysis Using Atom Search Optimizer and Ensemble Classifier

In modern days, a computer-based technique called natural language processing (NLP) is used to process spoken and written natural language data that is specific to human culture. Numerous NLP research directions and technologies have steadily arisen in the process of mining large amounts of textual data. Furthermore, prior models ignored the combined effect of relative distance and grammatical distance on the aspect words, instead only including the relative or grammatical distance of words into the model. This study examines the use of learning algorithms for Twitter sentiment analysis that is based on emotions like fear, anger, joy, and sadness. Sentiment-140 dataset is used in this research to conduct the analysis. First, a pre-processing step using tokenization, stemming, adding part of speech, and punctuation removal is carried out using the collected datasets. Then, using Bag of Words (BoW), Latent Dirichlet Analysis (LDA), Term Frequency, and Inverse Document Frequency (TF-IDF), the processed data is employed for semantic feature extraction. The next step is to choose the best features for accurate classification using the Atom Search Optimizer (ASO). Finally, the chosen features are divided into various classes using a Voting based Ensemble Classifier (VEC). From the experimental results, it is observed that the proposed VEC has outperformed existing methods by obtaining 98.69% of accuracy on and Sentiment-140 dataset.