Performance Evaluation of Multi-class Bangla Public Sentiment Analysis Using Machine Learning and Embedding Techniques

This study focuses on the use of embedding methods like word2vec and doc2vec in Bangla natural language processing (NLP) to represent textual documents as vectors or numerical data. These methods generate vectors for individual words and entire sentences or documents, optimizing computational resources and bandwidth. The study trains doc2vec and word2vec models on a corpus of 10,500 Bangla Facebook posts to classify sentiments as positive, negative, or neutral based on public reactions. Nine machine learning and deep learning techniques, including Logistic Regression (LR), Support Vector Machine (SVM), Decision Tree (DT), K-nearest Neighbor (KNN), Linear Discriminant Analysis (LDA), Gaussian Naive Bayes (GNB), Sequential Model (SM), Long Short-term Memory (LSTM), and Bidirectional Long Short-term Memory (BiLSTM), are used. The BiLSTM classifier achieves the highest accuracy, with a 77.34% accuracy through word2vec and a 75.56% accuracy by doc2vec . The LSTM model follows with a 76.71% from word2vec and 74.97% through doc2vec accuracy, respectively. The DT classifier performs poorly with the lowest accuracy scores of 57.9% and 51.02% for word2vec and doc2vec , respectively.