Recurrent Neural Networks with Support Vector Machines based Sentiment Analysis for Twitter Monitoring

In recent years, the sentiment analysis plays crucial role in area of Natural Language Processing (NLP) by identifying and classifying opinions expressed in text, especially on social media platforms like twitter. However, the existing Convolutional Bi-directional Recurrent Neural Networks (CBRNNs) faced challenges including poor performance on imbalanced data due to biased predictions and poor representation of minority classes. To address these limitations, this research proposes a hybrid RNN-Support Vector Machines (SVM) model by capturing long-term dependencies. Initially, the Twitter United States (TUS)-airline dataset consists of 14, 640 labeled tweets and preprocessed with multiple steps. After that, RNN is employed to extract temporal features from the tweet sequence which are then classified SVM by separates sentiment classes with optimal decision boundaries. This model enhances generalization, handles class imbalance more effectively and reduce training complexity. Experimental outcomes demonstrate that the RNN-SVM model by outperforms existing CBRNNs in term of accuracy (0.97), precision (0.99), recall (0.98), as well as F1-score (0.98), while monitoring social media platforms.