Analisis Sentimen dan Identifikasi Topik Ulasan Café di Malang dengan IndoBERT dan LDA

The rapid growth of the culinary industry in Malang has triggered fierce competition in the café business sector. Google Maps has become the main platform for consumers to leave reviews that affect business reputation. However, the large volume of diverse and varied reviews makes manual analysis difficult. This study aims to identify the sentiments expressed by consumers in reviews of cafes or coffee shops in Malang with a rating of 4.5 and above, as well as to discover topics of interest. The methodology combines IndoBERT for sentiment analysis and Latent Dirichlet Allocation (LDA) for topic modeling. The dataset consists of 71,073 reviews collected through crawling techniques. Testing was conducted in two scenarios: binary classification (positive and negative) and three-class classification (positive, negative, and other). The results show that the IndoBERT model in the binary scenario produced very high performance with 98% accuracy, while in the three-class scenario it produced 84% accuracy. Topic modeling with LDA through coherence score evaluation successfully identified two topics in negative sentiment, which focused on facility availability and service responsiveness. Meanwhile, seven topics were identified in positive sentiment, covering aspects such as taste, atmosphere, aesthetics, price, and social comfort. On the other hand, two topics in other sentiment tended to be neutral or had a mixture of positive and negative sentiments. This combination of methods proved to be effective in providing strategic insights for cafe owners to improve service quality based on customer feedback.