Neural Sentiment Network Model for Myanmar Language Using Attention Mechanism (AttenSentNet)

Sentiment analysis is the process of algorithmically identifying whether the opinion/emotion in social media posts, movie reviews as positive, negative, or neutral classes. It is becoming an important in making decision for business owners because they want to know the feedback of their products. It classifies the sentiment polarities into different classes in the user’s review. This proposed paper aims to predict the sentiment polarity of the user comment review from YouTube channels that are written in Myanmar language by combining BiLSTM (Bidirectional LSTM) and attention mechanism. Although there are many research works of NLP tasks using Myanmar language in machine translation and question answering systems, there are few researches on sentiment analysis tasks. The paper uses the hyperparameter tuning approach that is used to find the optimal parameters in training the neural sentiment model. The AttenSentNet also solves the vanishing gradient problem, which is a long-standing issue in the training the deep learning model. The attention mechanism pays attention to the important sentiment words in the given review. Experiments are tested on the phone product review in Youtube written by Myanmar language. Finally, the experimental results show with the accuracy of AttenSentNet model. To the best of our knowledge, this is the first work for sentiment analysis task on Myanmar language using BiLSTM neural network with attention.