SENTIMENT ANALYSIS OF MYANMAR NEWS AND COMMENTS USING SUPPORT VECTOR MACHINE

Abstract

As the development of internet technology is raising, the volume of information used for the internet users also increase in the web. Users can apply that information and give opinions for decision making system. Sentiment analysis also known as opinion mining is a task of text categorization methods that take opinion presented in a piece of text. An active research area is the sentiment analysis of text documents. The essential text resources found on social media, such as reviews, comments, tweets, posts, opinions, and articles, are available in a variety of languages. These could be analyzed to learn more about people's attitudes, beliefs, and feelings concerning various topics and products. With a focus on the Asian Language Treebank, news from Ministry of Information website(www.moi.gov.mm), and comments from social media webpage (www.facebook.myanmarcelebrity.com.mm), this paper aims to target news and comment of sentiment analysis in Myanmar social media. In order to categorize the sentiment polarity of each social media comment into "positive," "negative," or "neutral,”, automated analyzer methods were proposed in this paper. This system constructs corpus for news comments for Myanmar language. The datasets were then split into training and testing datasets, with the training dataset being randomly split in a non-overfitting way using the cross-validation approach. In order to improve the performance of the classifier, the case of imbalanced datasets was then considered. The hyperparameters were modified to improve the performance and outcomes of the classification. In addition, a number of information visualization techniques were used to display the results, indicate how effectively the classifiers performed, and highlight the key terms that had an impact on the classification process. Feature weighting and selection are required in sentiment analysis to get more efficiency. The proposed system implements sentiment analysis system for Myanmar News and comments. TF-IDF and N-gram are used for feature weighting and extraction. Support vector machine (SVM) is a supervised learning methods that analyze data and recognize the patterns that are used for classification. Hyperparameter optimization is used to find the set of specific model configuration arguments that does in the best performance of the model. Random search is an algorithm in which random combinations of hyperparameters are chosen and applied to train a model. The best random hyperparameter combinations are choosed. This system improves the Myanmar news sentiment analysis system using SVM with Random search optimization. This system also studies the machine learning algorithms for Myanmar sentiment analysis system. This system showed that the comparison results of Naïve Bayes, Linear SVC, and Linear SVC with random search optimization. Linear SVC with RandomizesearchCV has the highest performance. This system shows the most significant terms that had an impact on the classification process as well as the classifiers' performance. The results were then presented, along with ideas for how to optimize them in the further and information on how well the suggested systems worked.