Enhancing Sentiment Analysis on US-Based Twitter Data through Stemming in Long Short-Term Memory (LSTM) Networks

Sentiment analysis is a branch of Natural Language Processing which has been utilized by the organization and industries to identify the sentiments for the services and products. Sentiment analysis processed the textual data and convert semantics to identify the sentiment by dividing into three classes which are positive, negative and neutral. Here in this work, sentiment analysis has been examined using Long Short-Term Memory (LSTM) on twitter dataset. The data is preprocessed to convert into root words through the process of stemming for linguistic normalization and then into one-hot encoding, and embedding. The sentiment has been predicted using LSTM and the architecture has been systematically outlined with the precise value of the dropout layers to prevent the overfitting. The dataset features US airline tweets and addresses the real-world challenge of class imbalance in sentiment analysis. The model provides the robust training metrics, achieving an accuracy of 97.47%.