Multiple Approaches in Sentiment Analysis on Disneyland Reviews

Abstract

The digital transformation in society has greatly allowed people to have more freedom of speech. The volume for online opinion sharing is growing explosively and irreversibly. Hence, maximizing the utilization of these assets is the key to the success of a business, and sentiment analysis was introduced to conduct studies on the opinions. Sentiment analysis is a great tool to analyze and understand the needs of the customer. There exist multiple approaches in Sentiment analysis including Lexicon-based approach which uses a pre-trained model for unlabeled data, and Learning-based approach which builds a supervised machine learning model for labeled data. Furthermore, there exist multiple techniques to conduct sentiment analysis with these approaches. To estimate the performance of the different techniques, a case study is carried out which focuses on TextBlob and VADER for the Lexicon-based approach and focuses on SVM and Naive Bayes for the Learning-based approach. This study uses reviews posted by visitors on Trip Advisor for three Disneyland Resort Theme Parks. The results indicate that for the Lexicon-based approach, complete sentences with parameter tuning performs better than cleaned sentences without parameter tuning. It was found that VADER and TextBlob perform well on different theme parks. For the Learning-based approach, SVM performed better than the Naive Bayes technique. The Learning-based approach performed better compared to the Lexicon-based approach.