Indonesian Twitter Sentiment Analysis Using Word2Vec
Farhan Wahyu Kurniawan, Warih Maharani
Abstract
Sentiment is an opinion of someone on certain topics, products, or services. Sentiment analysis is used to analyze opinions to decide whether positive or negative. Twitter is used by Indonesian to express their opinion in the form of tweets. This study used word2vec method to extract features by converting data into vector. Word2Vec has the advantage of being able to see semantic relationship between words. This study used support vector machine (SVM) for the classification method. Sentiment classification process was done through processing training data in the form of tweets then turned into a model to be tested with the test data. First, tweets were collected by crawling them via Twitter API. Keywords and hashtags that are related to disasters in Indonesia were prepared to get relevance tweets for this study. For word2vec model, this study used Wikipedia in Bahasa Indonesia as a large amount corpus for model training. The training used Continuous Bag-of-Words (CBOW) and skip-gram as word2vec model architecture that in the last evaluation would be compared for the best classification result. Feature extraction was done by view each word in tweets to its vector score on the word2vec model that had already been trained. After that, the features passed to classification step. The implementation of word2vec and SVM generated the precision of 64,4%, recall of 58%, and an f-score of 61,1%.