Convolutional Neural Network for Predicting Sentiment: Case Study in Tourism

As a country with much natural wealth, Indonesia tries to utilize beach tourism in Bali to attract tourists. One of the websites in the tourism sector that is widely used by the world community today is TripAdvisor. Through TripAdvisor, tourists can find information about the beaches in Bali. Each beach has reviews written by tourists who have visited. However, reviews on TripAdvisor are unreliable and even biased. Therefore, Sentiment Analysis of Beach Reviews in Bali on the TripAdvisor Website can be a solution. This study uses real datasets from the TripAdvisor website in tourist reviews of the five most favorite beaches in Bali: Seminyak, Nusa Dua, Double Six, Kelingking, and Canggu. The research used the Convolutional Neural Network (CNN) architecture to produce positive and negative label predictions. The sentiment analysis results are visualized into a graph that describes tourist opinions on the five most favorite beaches in Bali. This study also measures the performance of the CNN model in making predictions. The accuracy obtained is 88% on Seminyak beach, 90% on Nusa Dua beach, 90% on Double Six beach, 87% on Kelingking Beach, and 85% on Canggu Beach. The CNN model performance measurement also produces precision, recall, and ROC curve on each beach.