Page 164 - Kỷ yếu hội thảo khoa học lần thứ 12 - Công nghệ thông tin và Ứng dụng trong các lĩnh vực (CITA 2023)

148

For data analysis in the hotel industry, researchers use a variety of approaches to
data analysis, ranging from statistical models to machine learning models [1, 12, 16,
19, 22, 23, 25, 26, 27]. However, to overcome the discretization of input variables and
output results, which can affect the accuracy of prediction, many studies have proposed
a long-term short-term memory (LSTM) to ensure the context of the sentence in the
process of data analysis [15, 17, 18]. Yazhou Zhang et al. [24] with the aim of
conversational sentiment analysis by adding a confidence gate before each hidden
LSTM unit to estimate the confidence of the previous speakers and combine the output
gate with the learned influence score to combine the influence of previous speakers.
Wei Li [13] proposed a lexically integrated two-channel CNN LSTM model that
combines the CNN and LSTM/BiLSTM branches in a parallel model to enhance the
accurate prediction of the sentiment polarizations of the user reviews. Results with
accuracy reach 94.6%.
Omran et al., [16] conducted using split test-validation training and k-fold cross-
validation to evaluate the performance of the model. They used measures of accuracy,
F1 score, and the AUC index. The model runs with average accuracy across all data
sets ranging from 96.72% to 97.04%, 97.91% to 97.93% at F1 score, while in AUC it
is 98, 46% to 98.7% when using reinforcement techniques.
Mohamed Arbane [17] experimented on the Twitter dataset where loss and accuracy
were used in the training and validation phases to evaluate the proposed Bi-LSTM
model. The obtained training accuracy was 95% and 86%.
In this study, we used an LSTM model with a 2-layer LSTM architecture to train
and predict customers' attitudes and emotions about Vietnam's hotel services with the
dataset that is collected from TripAdvisor.

3 LSTM

Long short-term memory (LSTM) is an artificial recurrent neural network (RNN)
architecture used in the field of deep learning. It was proposed in 1997 by Sepp
Hochreiter and Jurgen. LSTM is an improved network of RNNs to solve the problem
of remembering the long memory of RNNs. A common LSTM unit consists of a cell,
an input gate, an output gate and a forget gate. The cell memory values for an arbitrary
period of time, and the three gates regulate the flow of input and output information.
LSTM is well suited for classifying, processing, and predicting in the NLP field. The
structure of an LSTM is as follows:

CITA 2023 ISBN: 978-604-80-8083-9