Abstract Sentiment analysis is a field of study in natural language processing (NLP). This study proposes an approach to data processing, feature extraction, data balancing, and training using four machine learning models: Multinomial Naïve Bayes, Random Forest, Support... Sentiment analysis is a field of study in natural language processing (NLP). This study proposes an approach to data processing, feature extraction, data balancing, and training using four machine learning models: Multinomial Naïve Bayes, Random Forest, Support Vector Machine, and Decision Tree. Firstly, the dataset selected in the paper comprises the Internet Movie Database (IMDb), Twitter US Airline Sentiment (US Airline), and SemEval 2017. Second, data processing, feature extraction, and data balancing are employed to improve the accuracy of the training dataset. Specifically, data balancing is performed using the K-means SMOTE method, which has been proven effective for classification. Finally, the standard feature sets are applied to four machine learning models for training. The experimental results indicate that the SVM model achieves the highest accuracies of 89%, 96%, and 75% on the IMDb, US Airline, and SemEval 2017 datasets, respectively, compared to other state-of-the-art models. Read More Read Less

Abstract Dysphonia, a voice disorder characterized by abnormal vocal quality, significantly impacts communication abilities. Accurate and early detection is crucial for effective treatment and intervention. This study compares the efficacy of various Recurrent Neural Network (RNN) variants in... Dysphonia, a voice disorder characterized by abnormal vocal quality, significantly impacts communication abilities. Accurate and early detection is crucial for effective treatment and intervention. This study compares the efficacy of various Recurrent Neural Network (RNN) variants in classifying dysphonia using the Uncommon Voice dataset and provides an evaluation of standard RNN, Gated Recurrent Unit (GRUs) and Long Short-Term Memory (LSTM) models. Each variant was trained and tested on the preprocessed dataset, split into 80:20 ratio of training and testing sets. The finding shows variations in model performance, where the standard RNN achieved an accuracy of 76%, while the LSTM and GRU models demonstrated superior accuraci -es of 94% and 93%, respectively. These results underscore the potential of advanced RNN variants, partic ularly LSTM and GRU, for dysphonia detection and classification. The analysis offers preliminary information n about the relative advantages and disadvantages of each RNN variant, paving the way for future resarch in the broader domain of speech sound disorder identification. Read More Read Less