Sentiment Analysis with YouTube Comments Using Deep Learning Approaches

Social media had become popular over last decades and the users share their information to others, chats with their friends, families, and others about personal events, information, education and give their emotion with stickers, react signs, images and comments on social medias such as Facebook, YouTube, Twitters, Instagram via online platforms. So, using social medias between young, adult and old are highly improving after 2010. In Myanmar, most people use YouTube, especially, young and adult people use YouTube platform by uploading own created videos, short movie film, their favorite music videos, etc., share, and recommend feedbacks (comments) with their emotional feelings (happiness and hope, or sadness and angry). In Natural Language Processing, text classification is the most popular task. Sentiment analysis task is for extracting sentiment features in comments to classify positive, negative, or neutral. In the proposed system, over 9,000 comments are collected of the first trending No.9 Myanmar music video in YouTube, these comments are preprocessed including cleaning, stemming, segmentation, removing stop words, tokenization, and normalization. The main objective of the proposed system is to classify by extracting their sentiment features (only Myanmar Language) as positive, negative, and neutral using deep learning models Recurrent Neural Network (RNNs), Long Short-Term Memory (LSTM), Transformer Network, and Gated Recurrent Unit (GRU), compare the sentiment classification accuracy results and then system's performance is evaluated using confusion matric such as precision, recall, and F-score values. Finally, the proposed system showed that LSTM deep model is the highest with accuracy rate 97% precision 0.97, recall 0.96, and F-score 0.97 values among three models.