The development of digital platforms such as Google Maps has made customer reviews an important source of information for the hotel industry in evaluating service quality. However, the high volume of unstructured reviews and the presence of imbalanced sentiment classes make it difficult for management to identify problems and evaluate customer satisfaction quickly, accurately, and objectively. Therefore, this study aims to apply the Convolutional Neural Network (CNN-1D) algorithm to classify sentiments in Google Maps reviews of Hotel Grand Mansion Blitar, compare model performance before and after hyperparameter optimization using Grid Search based on Confusion Matrix evaluation (accuracy, precision, recall, F1-score), and map service priorities using Importance-Performance Analysis (IPA). The data were collected through web scraping using Webscraper.io, resulting in 1,580 raw reviews. After validation, text preprocessing, automatic sentiment labeling, tokenization, padding, and additional selection, 723 reviews were used in the modelling stage. Text representation was performed using Word2Vec. The results show that based on the Confusion Matrix evaluation, the baseline model achieved an accuracy of 66.21%, precision of 76.77%, recall of 66.21%, and F1-score of 70.45%. After optimization, the best model was obtained with 32 filters, kernel size of 3, dropout rate of 0.3, and learning rate of 0.001, achieving an increased accuracy of 81.38%, precision of 84.80%, recall of 81.38%, and F1-score of 82.92%. The IPA results indicate that the room condition (interior) attribute falls into Quadrant I, making it the main priority for improvement. Therefore, hyperparameter optimization significantly improved the sentiment classification performance, while IPA provided more focused service evaluation recommendations for hotel management.
Copyrights © 2026