Hyper-tuned Swarm Intelligence Machine Learning-based Sentiment Analysis of Social Media

Abstract

Natural Language Processing (NLP) uses Sentiment Analysis (SA) to determine text sentiment. SA is often used on text datasets to assess consumer demands, the sentiment of the customer for a product, and brand monitoring. Deep Learning (DL) is a subset of Machine Learning (ML) that mimics how humans learn. In this work, the Deep Learning Reptile Search Algorithm (SA-DLRSA) model is introduced for accurate automatic SA. The SA-DLRSA model utilizes Word2Vec word embedding to reduce language processing that is dependent on data pre-processing. The SA-DLRSA model utilizes SVM, CNN, RNN, BiLSTM, and BERT models for sentiment classification. Choosing the optimal hyperparameters is crucial for determining the model's architecture, functionality, performance, and accuracy. The Reptile Search Algorithm (RSA) is employed to find the best optimal hyperparameters to improve classification. A derived balanced dataset based on the tweets related to bitcoins was employed as a training dataset, which contains three sentiments, namely "neutral", "positive", and negative". The collection has 7 columns and 50058 rows, consisting of 21938 neutral, 22937 positive, and 5183 negative tweets. Precision, accuracy, recall, and F1 Score metrics were used to evaluate the effectiveness of the proposed approach. The results showed that the BERT and BiLSTM classifiers achieved superior performance in classifying sentiments in the tweets achieving accuracies of 99% and 98%, respectively. Due to the promising results of the proposed approach, it is anticipated to be used in solutions to social media problems, such as hate speech detection and emotion detection.