Visual Aspect-Based Sentiment Analysis for Tourist Reviews using Deep Learning Method

The tourism industry increasingly relies on user-generated visual content from platforms like TripAdvisor and Google Maps to aid decision-making and enhance customer experiences. This study focuses on analyzing images from reviews of major tourist destinations in Indonesia, such as Borobudur Temple, Prambanan Temple, Kawah Putih, and Mount Bromo. To assess the sentiments expressed in these images, we employ a Visual Aspect-Based Sentiment Analysis (Visual ABSA) model, which analyzes sentiment in a visual context—not only by examining the overall sentiment of the images but also by identifying specific aspects visible in the images and their accompanying sentiments using deep learning techniques, particularly Convolutional Neural Networks (CNN). Three architectures—ResNet-152, InceptionV3, and DenseNet-121—are evaluated for their performance. The results indicate that the DenseNet-121 model demonstrates strong and consistent performance in the location aspect, along with reasonably good results for price and atmosphere. Its superior performance is due to efficient feature reuse and dense connections between layers, enabling it to capture complex patterns with fewer parameters, which leads to improved accuracy.