A Lightweight Vision-Language Fusion Framework for Predicting App Ratings from User Interfaces and Metadata
The study proposes a lightweight vision-language framework to predict mobile app ratings using UI visuals and metadata. It combines MobileNetV3 for visual analysis of screenshots and DistilBERT for text analysis of metadata. A novel gated fusion mechanism, enhanced by the Swish activation function, merges these features to predict ratings with high accuracy.