Page 170 - Kỷ yếu hội thảo khoa học lần thứ 12 - Công nghệ thông tin và Ứng dụng trong các lĩnh vực (CITA 2023)

154

loss and accuracy for each iteration. Accordingly, to the 15th iteration, the accuracy
increases higher and the loss value decreases. This shows the effectiveness of the model
suitable for predicting customer sentiment towards hotel services in Vietnam.

Fig. 6. Loss and accuracy of LSTM model

We use Precision, Recall and F1-score to measures the effectiveness of the LSTM
model above. In which, it is found that the Precision measure has a result of up to 96%.

Table 2. The precision, recall and F1-score of LSTM model

Precision Recall F1-score
0 0.74 0.66 0.70
1 0.96 0.97 0.97
Accuracy 0.94
Macro avg 0.85 0.82 0.83
Weighted avg 0.96 0.94 0.94

To compare the accuracy of this model with the collected data set. We used
DecisionTree, RandomForest, LogisticsRegression and Kneighbors models, to
compare the accuracy with the LSTM model that we applied.

Table 3. Compare the used LSTM model with other machine learning models
Model Accuracy
Decision Tree 0.826277

RandomForest 0.860219
LogisticsRegression 0.889416
Kneighbors 0.859976
LSTM -ours 0.96

6 Conclusion

The trend of booking tours is growing, the number of travellers booking through online
channels is increasing rapidly, showing the fierce competition between hotels and

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