Hierarchical Sentence Sentiment Analysis Of Hotel Reviews Using The Naïve Bayes Classifier

• Published in: 2018 2nd International Conference on Informatics and Computational Sciences (ICICoS) pp. 1 - 5
• Main Authors: Kurniawan, Sandy, Kusumaningrum, Retno, Timu, Melnyi Ehonia
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.10.2018
• Subjects: Computational modeling; Data models; Feature extraction; Frequency measurement; hierarchical classification; Informatics; Naïve Bayes classifier; reviews; Sentiment analysis; Text analysis
• DOI: 10.1109/ICICOS.2018.8621748

Traveloka provides a space for its users to write reviews about their hotel reservation services. These reviews are very useful in informing hotel managers of the level of customer satisfaction. Sentiment analysis is a tool that can be used to analyse such reviews to determine whether they express opinions or not, so that the level of customer satisfaction can be measured based on the number of sentiments (positive or negative) contained in the opinions. In this research, the Naïve Bayes classifier was used to perform a hierarchical sentence sentiment analysis on hotel reviews obtained from Traveloka. In addition, two types of term weighting schemes were used for the feature extraction, namely, raw term frequency and TF-IDF. The results of this research indicated that it is better to use a hierarchical classification in sentiment analysis than a flat classification. The average F-measure value for the flat classification model was 75.18%, while for the hierarchical classification model it was 77.48%. These results showed that the use of a hierarchical classification in sentiment analysis improved the average performance of the classification model by 2.3%. The use of the raw term frequency feature extraction in a flat classification provided a higher F-measure value than the use of the TF-IDF feature extraction, with a margin of 3.9%. The average F-measure value for the flat classification using the raw term frequency feature extraction was 75.18%, while for the TF-IDF feature extraction it was 71.23%.