Performance Comparison of Word Embedding in Travel App User Review Sentiment Analysis

Abstract

Traveloka, as one of the leading travel booking platforms, has achieved more than 50 million downloads on Google Play Store. This achievement shows the high interest and trust of users in the services offered. However, user reviews indicate that there are some issues with the app's performance and stability that need to be taken into account. This research compares the performance of the Word2Vec and ELMo word embedding methods using the BiLSTM model in sentiment analysis of Traveloka application reviews. The research results show that the BiLSTM model with Word2Vec has an accuracy of 76.13%, precision 75.22%, and F1-measure 76.58%, better than the model with ELMo which has an accuracy of 74.38%, precision 70.49%, and F1-measure 74.40%. The BiLSTM model with Word2Vec is more effective in sentiment analysis of Traveloka reviews, helping identify and address user issues to improve service quality and user satisfaction.