Sentiment Analysis Using Natural Language Processing, Machine Learning and Deep Learning

Sentiment analysis is a vital branch of Natural Language Processing (NLP) that identifies the emotions expressed in a text. In this study, Long Short-Term Memory (LSTM) networks perform a unique sentiment analysis on data gathered from the Twitter site. This work intends to determine the performance of temporal models such as LSTMs in sentiment analysis tasks using many tweets from the Sentiment 140 dataset, which consists of 1.6 million tagged tweets. Following extensive pre-processing and creating the best possible LSTM model architecture, the model produced an astounding mean accuracy of almost 84 percent. Techniques like noise removal, tokenization, and removing stop words are used in pre-processing. Once the data is pre-processed, the required feature extraction is done using the GloVe embedding technique. Then, various machine learning models, like KNN, SVM, decision tree, and random forest, were built, along with a deep learning model using LSTM. These models are further evaluated using accuracy, precision, and recall metrics. The test set findings showed a considerable improvement, indicating a 15.53% increase in accuracy, a 16.56% increase in Precision, and a 13.31% increase in Recall compared to previously built Machine Learning models. This significant improvement demonstrates LSTM networks' capacity to detect and understand temporal relationships in textual data, confirming its efficacy in sentiment analysis. The study's conclusions demonstrate how LSTM networks can improve sentiment analysis tasks by offering a solid foundation for deciphering and interpreting subtle emotional differences in social media data.