<![CDATA[Customer reviews on Google Review contain important information about service quality, but the large and unstructured amount of data makes manual analysis difficult. This research aims to analyze the sentiment of customer reviews on Google Review Honda Malang Sekawan Motor Singosari using the IndoRoBERTa model thru a fine-tuning process. The dataset was obtained thru web scraping, followed by preprocessing which included cleaning, case folding, normalization, tokenization, stopword removal, and stemming. Sentiment labeling is performed using the lexicon-based InSet method as the initial labeling (weak labeling), then the dataset is divided into training and testing data with an 80:20 ratio. The model is used to classify sentiment into three classes: positive, neutral, and negative. The test results show that the model achieved an accuracy of 0.94, precision of 0.92, recall of 0.94, and an F1-score of 0.92. The model shows good performance in classifying positive sentiment, but still has limitations in recognizing negative sentiment due to data distribution imbalance. The contribution of this research is to provide an evaluation of the application of fine-tuning IndoRoBERTa in the domain of Google Review reviews of Indonesian-language automotive dealers, which have different characteristics from previous studies that mostly used social media data, while also analyzing the influence of InSet-based weak labeling and class imbalance on classification performance. The research results can serve as a basis for utilizing sentiment analysis to help dealers understand customer perceptions and identify aspects of service that need to be maintained or improved. The evaluation of the research also shows that the classification results still represent the label patterns generated by InSet, so further research needs to use human annotation as ground truth to improve the validity of the results.]]>
