43 Citations (Scopus)

Abstract

Product reviews are usually determined by sentiment of customers; however sentiment analysis based on aspects still need further research. A hotel commonly has five aspects, which are location, meal, service, comfort and cleanliness. This research proposes methods to determine review sentiment according to the hotel aspects. A hotel reviews are preprocessed into a term list. Firstly, Latent Dirichelet Allocation (LDA) determines the hidden topics of a term list; then Semantic Similarity categorizes the term list based on the topic resulted by Latent Dirichelet Allocation (LDA) into the five aspects of a hotel. Then in calculating similarity measurement, the term list is expanded by using the Term Frequency-Inverse Cluster Frequency (TF-ICF) method. Finally, a classification of customer sentiment (satisfied or dissatisfied) is conducted by using the combination of Word Embedding and Long-short Term Memory (LSTM). The results show that the proposed method can classify the reviews into the five hotel aspects. The highest aspect categorization performance is obtained by using LDA + TF-ICF 100% + Semantic Similarity which reaches 85%; the performance sentiment classification for the highest aspect-based sentiment analysis is obtained by using Word Embedding + LSTM which reaches 93%; and the comfort aspect receives more negative sentiments compared to the sentiments of other aspects. Also the results show that a sentiment is influenced by an aspect.

Original languageEnglish
Pages (from-to)142-155
Number of pages14
JournalInternational Journal of Intelligent Engineering and Systems
Volume12
Issue number4
DOIs
Publication statusPublished - 2019

Keywords

  • Aspect categorization
  • LDA
  • LDA-similarity
  • LSTM
  • Review analysis
  • Sentiment classification
  • TF-ICF
  • Term list
  • Topic modelling
  • Word embedding

Fingerprint

Dive into the research topics of 'Sentiment analysis of hotel reviews using Latent Dirichlet Allocation, semantic similarity and LSTM'. Together they form a unique fingerprint.

Cite this