<![CDATA[Internal company apps such as PT Astra Honda Motor’s AHM Mobile receive a wide range of reviews on the Google Play Store that contain users’ emotional expressions. However, the emotional dimension of these reviews has rarely been systematically analyzed, as most previous studies have focused only on positive, negative, and neutral sentiments. This study aims to analyze the emotions of AHM Mobile users through the integration of Natural Language Processing (NLP) and Computational Social Science perspectives. The dataset consists of 2.117 reviews obtained via web scraping and classified into six emotional categories: angry, sad, afraid, neutral, surprised, and happy. The annotation process was conducted by two annotators in the fields of clinical psychology and linguistics using a Seniority-Based Tie-Breaking mechanism with a Cohen’s Kappa value of 0.636. Emotion classification was performed using a combination of TF-IDF and Logistic Regression as classical models, as well as IndoBERT as the main model. Evaluation results show that the classical model achieved an accuracy of 0.43 and a macro-F1 score of 0.183, while IndoBERT reached an accuracy of 0.7831 and a macro-F1 score of 0.5582. Collective emotion analysis indicates that anger dominates user reviews and is strongly correlated with user ratings, as evidenced by a Spearman correlation coefficient of 0.7141. These results indicate that sentiment analysis using the IndoBERT model can provide more effective insights for evaluating the quality of usage of internal corporate applications in Indonesia.Keywords - Computational Social Science (CSS), Emotion Classification, IndoBERT, Natural Language Processing (NLP), User Reviews of the AHM Mobile Application.]]>
Copyrights © 2026
