Enhancing mobile app recommendations through adaptive fusion of long-term stability and short-term interests

The exponential growth in mobile application has greatly enhanced convenience in daily life, yet it has also complicated the process for users to find necessary apps in crowded mobile markets. This study presents an enhanced recommendation system that combines an attention-augmented Long Short-Term Memory (LSTM) architecture with a Bidirectional Recurrent Neural Network (BiRNN) to optimize user preference extraction. The attention mechanism enriches the traditional LSTM framework by prioritizing critical temporal data points, thereby improving both long-term preference processing accuracy and model interpretability. Experiments utilizing a dataset comprising 37,074 users and 52,783 applications demonstrated a 36.6% performance improvement compared to the Sequential Hierarchical Attention Network (SHAN) baseline model.