We developed an integrated framework to assess customer experiences at Buffet Poseidon by mining both textual reviews and photographs from Google Maps. In the textual component, user comments are classified into positive, neutral, and negative categories through a combination of traditional machine-learning techniques (e.g., logistic regression) and modern deep-learning architectures (BiLSTM, PhoBERT, Underthesea). Concurrently, visual data are processed by a YOLOv9-based image-classification pipeline, which assigns labels related to food presentation, portioning, and ambient features. The outputs of these two streams are then synthesized within a unified decision-support dashboard, enabling management to rapidly identify high-performing dishes and dining areas as well as those in need of adjustment. Although our methods achieve robust overall accuracy, the study also uncovers challenges associated with noisy inputs, computational limitations, and edge-case scenarios. These insights provide a practical, data-driven guide for targeted service enhancements in the restaurant context.