Sentiment Analysis Of Hotel Reviews On Tripadvisor With LSTM And ELECTRA

Abstract

This study examines the importance of hotel review data analysis and the use of Natural Language Processing (NLP) technology in predicting hotel review sentiment. In this study, deep learning models such as Long Short-Term Memory (LSTM) and Efficiently Learning an Encoder that Classifies Token Replacements Accurately (ELECTRA) are used to predict hotel review sentiment in Indonesian. Hotel review data was obtained through a data scraping process with webscraper.io from the Tripadvisor website and a total of 977 hotel review data were obtained from Grand Mercure Maha Cipta Medan Angkasa. Before the sentiment prediction process is carried out, hotel review data must go through the text preprocessing stage to remove punctuation marks, capital letters, stopwords, and a lemmatizer process is carried out to facilitate further data processing. In addition, sentiments that were previously unbalanced need to be balanced through the undersampling process. The data that has been cleaned and balanced is then labeled as negative (0), neutral (1) and positive (2) sentiments. The test results show that the ELECTRA model produces better performance than the LSTM with an accuracy of 47% by ELECTRA and 30% by LSTM.