Sentiment Analysis of Tourist Attraction Review from TripAdvisor Using CNN and LSTM

Abstract

The tourism sector has an important role in driving the economy. To find out the positive or negative responses of tourists, one of them is grouping through sentiment analysis using deep learning. The data used the tourist attraction dataset from TripAdvisor from several categories such as water and amusement park, nature, and museum. The methods used in this research are convolutional neural network (CNN) and long short-term memory (LSTM). In addition, Word2vec for feature extraction and Synthetic Minority Over-sampling (SMOTE) for handling imbalanced datasets will be used for this research. There are several scenarios used to perform sentiment analysis, with early stopping and with hyperparameter tuning using random search. The highest performance obtained on water and amusement park, nature, and museum category data is 83%, 97%, and 88% respectively for accuracy and 91%, 92%, and 93% respectively for F1-score. For the use of sentiment analysis methods, CNN can perform with the highest F1-score and LSTM can perform with the highest accuracy.

Exploring Sentiment Trends: Deep Learning Analysis of Social Media Reviews on Google Play Store by Netizens

“…Then there is research that performs sentiment analysis, such as sentiment analysis for hotel reviews using Aspect-Based Sentiment Analysis (ABSA) methodology [10], an aspect-level sentiment classification approach incorporating collaborative extraction hierarchical attention network [11], and sentiment analysis of campus teaching 2 regarding the implementation of Merdeka Belajar Kampus Merdeka utilizing naive Bayes and Euclidean distance methods [12]. However, many use LSTM in their sentiment analysis such as [13], [14], [15], [16], [17], [18], and [19]. In the study [20] several algorithms were compared and LSTM was chosen for its prediction stage.…”