Analysis of Word Vector Representation Techniques with Machine-Learning Classifiers for Sentiment Analysis of Public Facebook Page’s Comments in Myanmar Text

Abstract

This paper presents a study of comparison on three different machine learning techniques to sentiment analysis for Myanmar language. The fundamental part of sentiment analysis (SA) is to extract and identify the subjective information that is social sentiment in the source text. The sentiment class is positive, neutral or negative of a comment. The experiments are done on the collected 10,000 Facebook comments in Myanmar language. The objective of this study is to increase the accuracy of the sentiment identification by using the concept of word embeddings. Word2Vec is used to train for producing high-dimensional word vectors that learns the syntactic and semantic of word. The resulting word vectors train Machine Learning algorithms in the form of classifiers for sentiment identification. This experimental results prove that the use of word embeddings from the collected real world datasets improved the accuracy of sentiments classification and Logistic Regression outperformed the other two ML methods in terms of accuracy and F-measures.