Topic and Sentiment Classification of Streaming Tweets about Tourist Destinations in Thailand

Abstract

In this research, a website about Thailand's tourist destinations was implemented in a responsive style to support both desktop and mobile displays. It retrieved live streaming tweets about specified destinations and classified them by topic (into News, Foods, Environment, or Traffic) and by sentiment (into Positive, Negative, or Neutral). Using recent state-of-the-art Word2Vec embedding, along with support vector machine classifier, the accuracy of topic classification was 80% and that of sentiment classification was 59%. In addition, based on the website evaluation by 30 users, an average satisfaction score of 4.4 out of 5 was achieved.